1979 , Volume , Issue Jan-1979

1979 , Volume , Issue Jan-1979
HEWLEITPACMRD JOURNAL
JH:
r
© Copr. 1949-1998 Hewlett-Packard Co.
A Low-Cost, Microprocessor-Based,
100-MHz Universal Counter
A special integrated-circuit counter chip works with the
microprocessor to give this reciprocal-taking counter a
range of capabilities formerly found only at a much higher
price. Flexible input amplifiers, a novel battery pack, and
low radiated emissions are other features.
by Lewis W. Masters, Karl M. Blankenship, and Michael J. Ward
MICROPROCESSORS AND LARGE-SCALE inte
gration (LSI) have become powerful and indis
pensable tools for the designer of measurement in
struments, allowing as they do a reoptimization of the
conflicting goals of performance, features, and cost.
Until now, these tools have made maximum impact
at opposite ends of the instrument cost spectrum.
Microprocessors have improved performance in
high-end instruments ranging from voltmeters to
spectrum analyzers, and LSI technology has provided
simple multimeters and frequency counters at very
low cost.
In HP's new 5315A/B Universal Counter, these two
design tools have been combined to advance the state
of the art in the middle range of counters by providing
improved performance, more features, and reduced
complexity at a much lower price than previous gen
erations. Three ICs are critical to the performance of
the 5315A/B: a commercial single-chip microcomput
er, an integrated display driver chip, and most impor
tant, an HP-made custom LSI chip containing all of
the data-acquisition functions needed by a universal
counter. The implementation of this last 1C, which we
call the multiple-register counter, or MRC, is de
scribed in the article on page 12.
The 5315A/B is a moderately priced universal
counter that accepts signals from dc to 100 MHz on
both input channels. Its basic time interval resolution
is 100 nanoseconds, but this may be improved on
repetitive signals to less than a nanosecond through
the use of averaging. The 5315A (Fig. 1) comes in a
rugged plastic case that is suitable for bench use and
can be equipped with an optional internal battery
pack for portable and field applications. The 531 5B is
the same instrument housed in a standard HP metal
package. The 5315B is useful for system applica
tions or where superior EMI (electromagnetic inter
ference) performance is required. Either instrument
may be ordered with a temperature-compensated
oscillator for applications where higher accuracy
is necessary.
A Computing Counter
The primary feature distinguishing the 5315A/B
from earlier moderately priced counters is that it is a
computing or reciprocal frequency counter. Simply
put, the 5315A/B measures the period of the input
signal and computes the corresponding frequency
(frequency = I/period) for display. This may seem
Cover: Model 531 5 A Uni
versal Counter is surrounded
by microphotographs of
the MRC (multiple-register
counter) chip, a state-of-the
art HP-developed bipolar
integrated circuit designed
to perform most universal
c o u n t e r f u n c t i o n s u n d e r m i
croprocessor control. The MRC is one of three
major ICs in the 531 5 A.
In this Issue:
A Low-Cost, Microprocessor-Based
100-MHz Universal Counter, by Lewis
W. Masters, Karl M. Blankenship, and
Michael J. Ward .
page 2
Lowest-Cost HP Universal Counter Developed
Using LSI and Manufacturing Innovations, by
Michael D. Wilson and David M. George, pages.
A High-Performance Bipolar LSI Count
er Chip Using EFL and I2L Circuits, by
Sosco W. Wong and William D.
J a c k s o n
page 12
A Synthesized Signal Source with
Function Generator Capabilities, by
Dan D. Danielson and Stanley E.
F r o s e t h
page 18
Viewpoints — Paul Baird on Electronic
Equipment Reliability
page 27
Printed in U.S.A.
Hewlett-Packard Company 1979
© Copr. 1949-1998 Hewlett-Packard Co.
Fig. 1. Model 5315A Universal
Counter measures frequency to
100 MHz and time intervals with
100-ns resolution. A reciprocaltaking two-channel instrument
with a seven-digit display, it offers
mid-range counter performance at
a cost much lower than previousgeneration counters. Shown here
is the 531 5A. The 531 56, thesame
instrument in a half-rack-width
metal cabinet, offers exceptionally
low radiated emissions.
like unnecessary complexity, since a conventional
direct-gating counter provides high resolution at
communications frequencies. However, the conven
tional counter is restricted to 1-Hz resolution per sec
ond of gate time, so at low frequencies, it often pro
vides unacceptable results. For example, it would
take a 100-second gate time for a direct counter to
provide a one-part-per-million accuracy check on a
32-kHz watch crystal. The high resolution of the
computing technique enables the 5315A/B to display
at least seven digits of answer per second of gate time,
independent of the input frequency. Therefore, it
would take only 0.1 second to measure 32 kHz, or for
that matter 32 Hz, to one part per million.
channel is enabled. This feature allows time mea
surements to be made on noisy signals or where sig
nal anomalies prevent the standard function from
operating properly (Fig. 2).
In the ratio mode, the ratio of the frequency in
channel A to the frequency in channel B is displayed.
Two types of totalizing functions are provided, one
manually controlled by the operator and the other
_TL
A Input
B Input
Stop Enable
Universal Measurement Capability
Besides high-resolution frequency measurements,
the 5315A/B can make many other kinds of single and
dual-channel measurements. The period measure
ment capability of the counter is virtually identical to
the frequency capability, since the only difference
between the two is how they are displayed. Three
types of two-channel time measurement functions are
offered: time interval, time interval average, and time
interval with delay. All three functions provide time
measurement from an event at the A input to an event
at the B input and provide at least 100-ns resolution.
Time interval average allows greatly increased mea
surement resolution on many repetitive signals. Time
interval with delay allows the user to vary the time
between the start event and the time when the stop
Delay Time
Measured
Times
Example of Use of Time Internal Delay Function
(Slope Controls are Both Positive Edge)
T, is the Result if DsT,
(It Is Also the Only Result a Standard Tl Function Will Give)
T2 Is the Result if T, D- T2
T3 Is the Result if T2- D- T3 Etc.
Fig. 2. Time interval with delay is useful on noisy signals or
signals with anomalies.
© Copr. 1949-1998 Hewlett-Packard Co.
controlled by electrical events applied to the second
input. Several auxiliary functions are available, in
cluding the traditional counter check function, a dis
play verification function, and a way to measure the
gate and delay times. A position on the function selec
tor (FREQ C) has been reserved for a frequency extention option to be introduced soon.
Display Features
A number of steps were taken to enhance the read
ability of the display. First, yellow LEDs were
selected because they are highly readable and easy on
the eye. Second, the result is displayed in engineering
format, with eight digits reserved for the mantissa and
one for the exponent. Engineering format eliminates
the confusing array of annunciators usually found on
counters and replaces them with exponents of 0, ±3,
±6 and ±9.
The most-significant digit is always placed in the
leftmost LED so that the operator knows where to
look. Similarly, the decimal point may appear in only
the three leftmost digits, eliminating correct but con
fusing displays such as 32128 kHz.
Most measurement functions will not produce a
display overflow; only the eight most-significant dig
its are shown. Totalize and frequency, however, may
contain useful data in the least-significant digits, so
when more than eight digits of result are available,
the most significant digits are overflowed. For in
stance, without overflow, the 5315A/B would be lim
ited to 10-Hz resolution at 100 MHz, but with it,
0.1-Hz resolution is available.
The display area also contains annunciator LEDs
for the optional battery pack and an error LED that
turns on if the microcomputer finds an internal error
during its start-up check.
Fig. 3. The 5315A/B block diagram is similar to that of a
computer system. A commercial microcomputer controls a
data acquisition block — a multiple-register counter chip —
and the display.
timize resolution and update rate. With short gate
times the display is updated rapidly, making it possi
ble to follow changing signals. Long gate times pro
vide more digits in the result. When the approximate
measurement time must be known, selection of the
GATE TIME function will display the current setting.
The gate-time control sets approximate measurement
time in frequency, period, time interval average,
check and ratio functions. In time interval measure
ments with delay, the control varies the holdoff time
before the stop is enabled, and in standard time inter
val measurements acts as a sample rate control. When
the knob is rotated to the HOLD position, the counter
makes one measurement at minimum gate time and
then halts. A new measurement may be initiated by
pressing the reset button.
Continuously Variable Gate Time
The 5315A/B is the first counter to offer a continu
ously variable measurement time control. This fea
ture is possible because a computing counter is not
constrained to a fixed number of input events, such as
100, to yield a correct display. The microcomputer
solves the equation: Frequency = (number of periods,
N) -T- (total time, T, for N periods). N and T can be any
number greater than 1 and less than 1012.
The front-panel gate-time knob varies the time con
stant of a resistor-capacitor timer that determines the
time from the start of a measurement until the stop is
enabled. The next input event terminates the mea
surement. The exact measurement time is not impor
tant so long as it is known precisely enough. The
counter measures this time to an accuracy of 100
nanoseconds.
A continuously variable control is useful on a
counter because it gives the user a quick way to op-
Computer-Type Architecture
The block diagram of the 5315A/B should be quite
familiar, since it is the same as that of a classical
computer system (Fig. 3). There is a data-acquisition
block (the MRC), a computation and control block
(the microcomputer), and an output block (the dis
play). LSI technology has reduced each of these three
blocks to virtually a single 1C. The MRC requires only
a reference oscillator (time base) and a half-dozen
components to do all the gating and counting re
quired by a universal counter. The MRC was designed
specifically as a microcomputer peripheral, so it in
terfaces directly, with a minimum of lines and no
other components. The microcomputer, a 3870 with
2K bytes of on-board read-only memory (ROM), is
likewise a model of simplicity, consisting of only a
single 1C and a few discrete components. At one stage
4
© Copr. 1949-1998 Hewlett-Packard Co.
in the development of the 5315A/B, the display con
tained more components than any other block, but
later an 1C manufacturer was found who had the
capability and desire to integrate a complete LED
drive subsystem. The result, Intersil's ICM 7218A, is a
microcomputer peripheral chip that requires no other
components to drive eight standard-size LED digits.
The simplicity of these three blocks is also mirrored
in the rest of the 5315A/B. In all, only a dozen ICs were
required to implement this counter, about 30% of the
number required to build a predecessor of much
lower capability.
The interaction of the various functional blocks to
make a measurement is straightforward, with the
microcomputer providing all instrument control via
data, address, and command buses that use 16 of its
general-purpose I/O ports. On power-up, the mi
crocomputer goes through a brief check routine to
verify its ROM, RAM, and I/O ports. It then checks to
see that the MRC can make a simple measurement. If a
fault is detected, a front-panel error LED is turned on
and an error code displayed briefly. Next, the com
puter checks the front-panel pushbuttons to deter
mine what function is selected. The appropriate set
up data for that type of measurement is then sent to
the MRC control register, an operation analogous to
writing to a RAM (random-access memory). The mi
crocomputer next resets the MRC's counting and gat
ing circuits and enables the start of a measurement
cycle.
From this point on, the computer has nothing to do
until signaled by the MRC that service is needed. The
MRC signals the computer whenever it has completed
a measurement or one of the eight decade counter
chains has overflowed. The computer can then access
the MRC status register to determine the proper
course of action. The computer must know when an
overflow occurs because the counter chains are ex
tended another four decades in the computer's mem
ory. When the microcomputer detects the end of a
measurement, it reads data from the two MRC count
ing registers, one BCD digit at a time.
Once the computer has the data, it processes it into
a usable form, usually by dividing one 12-digit data
word by the other. The result of this division, prop
erly scaled, is usually the data that is displayed. First,
however, the microcomputer manipulates the
number to give a meaningful display. Even though
the result of the division is a 12-digit number, very
few measurements produce results with that resolu
tion. Therefore, the number is usually truncated so
that only statistically significant results are dis
played. However, it is important to display as many
digits as possible, so the truncation algorithm must be
versatile and accurate. The microcomputer examines
the original data from the MRC to determine which
digits are significant, then shifts and manipulates the
data so that it is most useful to the user when output to
the display subsystem. The computer sends data to
the display driver 1C in much the same way that it
would to a RAM. Once the data is in the display, the
microcomputer is free to start a new measurement
cycle.
External Control
Although the microcomputer has a limited memory
capacity, some memory was left over after the basic
control and computation routines were completed, so
a useful routine was written to fill the remaining
program space. This routine allows a user, through an
optional programmer box, to do mathematical ma
nipulation of the data displayed by the counter. With
this offset/normalize option, the counter can display
the solution to the equation x/m + b, where x is the
normal result and m and b are variables input by the
user on thumbwheel switches. The data is input in
scientific notation with eight digits plus sign for the
mantissa and one digit plus sign for the exponent.
When a negative answer is displayed, the leftmost
LED digit becomes a minus sign, reducing display
resolution to seven digits. Display range is from
999. xlO9 to less than .lxlO~9. The offset/normalize
option is useful in transducer measurement applica
tions, where output is desired in such units as revolu
tions per minute or gallons per hour instead of events
per second.
Input Amplifiers
The input amplifier portion of the 5315A/B Counter
takes the unknown signals, which can have a wide
range of frequencies and amplitudes, and amplifies,
conditions, and converts them into logic-level signals
that can be counted by the MRC chip. The nominal
sensitivity of the amplifiers is 10 mV from dc to 10
MHz, with a slow rolloff in frequency response above
10 MHz to a nominal sensitivity of 25 mV at 100 MHz.
These specifications apply to both channels, A and B.
The B-channel amplifier is matched to and is substan
tially identical to the A-channel amplifier, making
possible such measurements as 90-MHz frequency
ratios and time interval averaging to better than 5-ns
accuracy.
Front-panel controls include ac/dc coupling and a
X20 attenuator for each channel, a 100-kHz low-pass
filter on the A channel, a separate/common-A switch,
individual slope select and trigger level controls, and
a novel feature, a sensitivity control mode that is
individually selectable for each channel. Trigger
lights indicate whether the input signal is being cap
tured by the counter, and in dc coupling, provide
information about the relationship of the input signal
level to the trigger level.
The amplifier is constructed and serviced as a sepa-
© Copr. 1949-1998 Hewlett-Packard Co.
Comparator
Channel A
Input
X1
-K>-
In
dc
Channel A
Trigger Level/
Sensitivity
Control
Channel B
Input
Channel B
Trigger Level/
Sensitivity
Control
Trigger Level
Sensitivity
Switch
Slope B
To Microcomputer
Fig. 4. 5315A/B input amplifier
section.
rate module. Amplifier modules are interchangeable:
any amplifier will work with any 5315A/B main
board. This feature improves flow in the manufactur
ing area and simplifies service.
The functional block diagram of the amplifier sec
tion of the 5315A/B (Fig. 4) is straightforward. A dual
high-speed comparator provides most of the amplifi
cation. The dc trigger level is applied to the reference
input of the comparator and the input signal is de
coupled to the other input. The comparator output, at
standard ECL levels, is shifted to the special levels
required by the MRC.
The trigger level is controllable separately for each
of the two channels. In trigger level mode, the trigger
level potentiometer applies a variable voltage ( + 2 to
-2 Vdc) to the reference input of the comparator. This
voltage defines the threshold dc level at which the
output of the comparator switches states.
Sensitivity Mode
A new feature in this counter is the sensitivity
mode of operation. This is selected by pushing the
button next to the trigger level potentiometer. In this
mode the trigger level presented to the comparator's
reference input is set to ground potential, so that the
comparator is now operating as a zero-crossing detec
tor. The trigger level potentiometer does double duty,
in this mode operating as a sensitivity control.
In sensitivity mode the trigger level potentiometer
controls the value of the comparator's hysteresis and
thereby controls the threshold switching voltages.
This effectively controls the sensitivity of the
amplifier; in this respect the potentiometer may be
thought of as an amplifier gain control. When the
trigger level control is fully clockwise, the sensitivity
is maximum. Turning the knob counterclockwise in
creases the hysteresis and decreases the sensitivity.
When the knob is turned fully counterclockwise, the
amplifier is essentially turned off. In the trigger level
mode, the sensitivity is automatically preset to
maximum.
In general, the sensitivity mode works best for most
frequency measurements and the trigger level mode
for most time interval measurements. However, there
are exceptions.
For low-frequency signals it is necessary to use dc
coupling. In this case, the trigger level mode is gener
ally used, even for frequency measurements. How
ever, sine waves with no dc offset cause no trouble in
de-coupled sensitivity mode.
With signals larger than 100 mV rms and the
counter in sensitivity mode, the sensitivity control
may be set somewhat counterclockwise (reduced sen
sitivity) if it is suspected that the input signal has
noise riding on it. This allows the input amplifier to
ignore the smaller noise impulses while still trigger
ing properly on the desired signal. This does not help,
of course, if the noise amplitude is greater than that of
the desired signal. In this case, if the input signal is
below 100 kHz and the interfering noise is highfrequency, using the A-channel filter will help sup
press the miscounts due to noise. In other cases it may
© Copr. 1949-1998 Hewlett-Packard Co.
10-A PNP V T
Pass Transistor
To Load and
h3-Volt Regulator
Battery or
Unregulated dc
from Line
Fig. 5. +5V regulator uses a PNP pass transistor instead of
the usual NPN type. Dropout voltage is only 200 mV instead of
the usual 2V, making it possible for the 6V battery pack to
supply the +5V and +3V regulators directly, without dc-dc
conversion.
be necessary to filter the input signal before it is
applied to the 5315A/B inputs.
In measurements of impulses or pulse trains, the
trigger level mode is usually best. This mode allows
the user to select where on the pulse waveform the
counter will trigger, and to choose triggering above or
below noise or ringing at edge transitions. Alterna
tively, the user may choose to trigger during the ring
ing, to characterize it. If very large signals are in
volved, the x20 attenuator can be switched in.
Power Supply
The 5315A can be operated either from the ac
power line or from an optional 6-volt battery pack.
The 5315B operates from the ac line only.
For ac line operation, 100/120/220/240 volts at
48-66 Hz is applied to the ac portion of the power
system, consisting of transformer, rectifier bridge,
and filter capacitors. This section produces the posi
tive and negative unregulated dc voltages. The
unregulated dc voltages are routed via the power
switch to the regulator section of the circuit, which
provides the + 5V, + 3V, and -5.2V supply voltages
necessary to run the counter. All three regulators are
adjustable.
The adjustability of the regulators has certain ad
vantages in the production of a modular instrument
such as the 5315A/B. Being able to set each voltage
accurately allows the design of certain modules to be
optimized with respect to the supply voltages. This
means that the functional modules can be tested
while separate from the mainframe, with a high level
of confidence that subsequent mating of the modules
with the mainframe will yield optimum performance
without additional testing and adjustments.
The design of the +5V regulator makes it possible
to use only a 6V battery in the battery pack, thereby
reducing the size, weight and cost of the battery pack
significantly compared to an 8V battery, the next
available size. The -I- 5V regulator is a discrete series
regulator with an unusual feature. It employs a highcurrent PNP pass transistor instead of the usual NPN
type (Fig. 5). This change offers significant advantages
when operating from the battery pack, because it
allows operation down to an input-output voltage
differential (drop-out voltage) equal to the Vce(sat) of
the transistor, or less than 0.5V. By comparison, the
dropout voltage of a typical three-terminal regulator
is about 2V.
The Vce(sat) of a transistor is minimized when the
transistor is operated at a current level much lower
than its rated capability. The transistor selected for the
+ 5V regulator allows the circuit to regulate down to an
input-output voltage differential of approximately 200
mV. This means that a 6V battery can be used to supply
the +5V and +3V regulator chain directly and dc-dc
conversion is necessary only to supply the negative
supply voltage. Because most of the power in the in
strument is on the -I- 5V and + 3V buses, the system is
very efficient. In battery operation, the + 5V regulator
is over 75% efficient. Had a three-terminal regulator
and an 8V battery been used, the efficiency would have
been 15-20% lower.
Battery Pack
The optional 5315A battery pack consists of the
6-volt rechargeable battery and the circuitry neces
sary to charge and discharge it. It also incorporates a
dc-dc converter (inverter), which supplies the nega
tive voltage for the mainframe during battery opera
tion. The battery pack is designed to provide adequate
power to run the instrument and yield optimum per
formance in applications ranging from standby oper
ation to once-a-day cycling. Special attention is also
given to maximizing battery life through features that
minimize the effect of abusive use and provide useful
information to the user.
The battery selected for the 5315A consists of three
series-connected, 2V, five-ampere-hour Gates re
chargeable sealed lead-acid cells. Lead-acid cells
offer high cell voltage, low cost, long shelf life, and no
discharge memory.
Current-limited constant-voltage charging and
constant-current charging are two common tech
niques for charging lead-acid batteries. Neither of
these techniques offered a solution to the charging
requirements of the 5315A battery pack. To recharge a
battery of this capacity in a time period suitable for
once-a-day cycling using the current-limited
constant-voltage technique requires either a very
© Copr. 1949-1998 Hewlett-Packard Co.
Lowest-Cost HP Universal Counter
Developed Using LSI and Manufacturing Innovations
by Michael D. Wilson and David M. George
Recent advances in LSI technology have produced a univer
sal counting 1C with the oscillator circuit and display driver
incorporated into the design. Previously, this much circuitry
required one or more printed circuit boards, packed with small
and medium-scale TTL integrated circuits.
This new 1C, the Intersil ICM 7226A, is used in a new low-cost
HP universal counter, Model 531 4A (Fig. 1). The 531 4A is a
100-MHz universal counter that measures frequency, period
average, time interval, and ratio in conventional fashion. It also
has totalize and check functions.
Fig. 2. The only two boards in the 5314A are made, loaded,
and tested as a single unit, thereby lowering manufacturing
costs. The boards are separated before being installed in an
instrument.
Fig. 1. Model 5314A is a low-cost two-channel 100-MHz uni
versal counter that measures frequency, period average, time
interval and ratio.
The 531 4A is the lowest-priced universal counter manufac
tured by HP. To achieve a low price, the instrument and power
supply (the only boards in the standard 531 4A) are loaded and
tested as a single printed circuit board (Fig. 2). The only adjust
ment required is the tuning of the 10-MHz reference oscillator,
and this can be done from the front panel of the standard
instrument. The front board and metal front panel are held in
place an plastic inserts molded into the package, resulting in an
instrument that is easy to assemble and requires no expensive
hardware.
Low cost was achieved while retaining performance by de
signing both technically efficient circuits and an equally efficient
manufacturing plan. The simple notion of building the complete
instrument on a single printed circuit board has resulted in cost
savings in every step of the manufacturing process. The single
blank board is cheaper than two separate boards, the loading of
parts test quicker, secondary operations are eliminated, and test
ing is man In general, one part passing through the man
ufacturing system is more efficient than two smaller parts. Extra
emphasis was placed on the physical board layout to minimize
the possibility of misleading parts. With a low-cost instrument,
one cannot afford to spend much time correcting errors. There
fore, many parts are automatically inserted under computer
control. Both custom and standard resistor packs are used and
special provisions made so they cannot be loaded incorrectly.
Since testing can be very time-consuming, it was essential
that the 531 4A take advantage of the most efficient testing
methods available. The instrument is built on a single flat printed
circuit board, making it possible to test it quickly with a vacuum
fixture and an HP 3060A Board Test System. Accurate diagnos
tics for units that might have a problem eliminate tedious
troubleshooting.
The 531 4A uses yellow seven-segment LED displays instead
of red. Yellow has better contrast at distances greater than eight
feet, of unlike red, is visible to people with certain types of
color blindness. Custom resistor packs were designed to obtain
short electrical paths, significantly improving high-frequency
performance while simultaneously lowering parts count. This
helped to provide flat sensitivity of 25 mV to 100 MHz.
ECL line receivers are used for amplification because of their
low cost and high gain at high frequencies (100 MHz). This
allows the internal power supply design to be simplified be
cause only +5 volts is required. Trigger levels and slope con
trols can be set from the front panel, enabling the counter to
function correctly with all waveforms. A separate/common-A
switch allows timing measurements to be made using only
Channel A or both channels. For frequency measurements the
counting chain will direct count to 10 MHz, and an ECL prescaler can be switched in to allow counting of frequencies to 1 00
MHz. To conserve power, the prescaler is disconnected when
not in use, reducing power consumption by more than 15%.
Two options are available: a temperature-compensated crys
tal oscillator (TCXO) and a battery pack. The battery is automat
ically switched in and will power the instrument for about eight
hours any time ac power is not present. The battery pack and
charging scheme used in the 531 4A are similar to (but simpler
than) those of the 531 5A Universal Counter, as described in the
accompanying article.
A comparison of the 531 4A with a predecessor, the 5300B/
5302A, shows that the 531 4A actually has superior capability in
frequency response, sensitivity, and timing resolution, yet sells
© Copr. 1949-1998 Hewlett-Packard Co.
for less than half the price. Not only does the 531 4A provide
good performance, it should be very low in maintenance re
quirements due to its high estimated mean time between fail
ures (MTBF = 40.000 hr) and easy access to the printed circuit
boards. No extender boards or special cables are required.
A special power transformer conforms to IEC and VDE (Verband Deutcher Electrotechniker) safety standards. The stan
dard 531 4A has exceptionally low electromagnetic emissions.
Extensive environmental and life testing insure that this instru
ment of provide good measuring capability in a wide variety of
applications.
Acknowledgments
Without the involvement of many people this project would not
have been possible. Both Jim Horner and Ian Band provided
valuable criticism and encouragement to keep the project on
David M. George
David George received his BSEE
degree from the University of Utah
in 1976 and his MSEE from Stan
ford University in 1977. Joining HP
in 1977, he took the 531 4A Univer
sal Counter from initial design
through its release to production.
David was born in Waco, Texas
and grew up in Ogden, Utah. He is
single, lives in Campbell, Califor
nia, and enjoys skiing, skin diving,
and hiking.
high current level or a very high float voltage level.
The required current level exceeded the power limita
tions of both the power transformer and the instru
ment package. The high float voltage would severely
reduce the life of the battery in standby applications.
While the constant-current technique would provide
recharge times suitable for once-a-day cycling at a
moderate current level, problems exist in sensing
end-of-charge reliably at that current level. This
could result in reduced battery life due to overcharg
ing and/or less than optimum performance in standby
applications.
The charging circuit used in the 5315A battery pack
takes advantage of two characteristics of the Gates
lead-acid cell. When charged at a constant current
equal to one-tenth of the rated capacity, the cell volt
age reaches a particular value after 90% of the charge
previously taken from the cell is returned. Also, in
an overcharge state, application of a constant current
causes the cell voltage to float near a particular level.
Using these two characteristics, it was possible to
design a charger capable of recharging a fully dis
charged battery in 16 hours, reliably sensing end of
•The into figure does not take the charge acceptance of the battery into account. If the charge were
terminated at this point and the battery subsequently discharged, approximately 72% of the rated
capacity would be available
target for a timely introduction while meeting the technical goals
Valuable mechanical contributions were provided by Eric
Havstad, Ashok Phadke and Mike Detro. These people de
veloped several previously untried ideas, many of which were
incorporated into the final design. Bob Bliven provided many
good ideas on the front-panel layout in spite of countless
changes. Lew Masters, with his previous experience in low-cost
counters, was helpful in product definition. Ivan Andres also
provided good support and many good ideas for this project.
Numerous production people provided useful information to
make the design more producible.
Michael D. Wilson
Born in Monterey Park, California,
Mike Wilson is a 1972 BSEE and
1973 Master of Engineering
i graduate of California State
I Polytechnic University at San Luis
1 Obispo. Mike joined HP shortly
after graduation in 1973 and has
worked as design engineer on the
5328A Universal Counter and as
project leader on two 5328A op
tions and the 531 4A Universal
_ Counter. A member of IEEE, he is
-•^^ currently production engineering
.. .^^^gEP- - / supervisor for logic test and preml cisión frequency sources. Mike is
married, has a 1 9-month-old daughter and lives in Santa Clara,
California. In the summer months, he enjoys catamaran racing
and playing softball, while his winter months are spent skiing
and playing volleyball.
charge, and providing a safe float voltage for standby
applications. It accomplishes this at moderate power
and good efficiency. The charging technique is called
"two-step constant-current state-of-charge sampled,"
or in fewer words, hysteresis charging.
The circuit consists of a parallel arrangement of
high and low-current regulators that supply charge
current to the battery, and a voltage comparator cir
cuit with hysteresis that monitors the battery voltage
and controls the operation of the high-current reg
ulator. The charging cycle involves three stages of
operation (see Fig. 6]. The battery is initially supplied
with a constant current of 500 mA from both current
regulators (high current = 490 mA; low current = 10
mA). The upper threshold of the comparator hys
teresis window coincides with the battery voltage
corresponding to the 90% charge-returned point.
When the battery voltage reaches this level, the com
parator disables the high-current state, and the bat
tery voltage starts to drop. The lower threshold of the
comparator is set below the overcharge float voltage
level corresponding to the 10 mA constant current.
When the battery voltage reaches this lower
threshold, the comparator enables the high-current
regulator and the charge current returns to 500 mA.
The battery voltage again rises to the upper threshold
© Copr. 1949-1998 Hewlett-Packard Co.
High
Current
the battery at high and low temperatures. This is
accomplished by automatically adjusting the level
and size of the charge comparator hysteresis window
as a function of temperature.
In battery operation the 5315A battery pack pro-
Charge Current
Low
Current
Time=16 hr
Battery Voltage
Lewis W. Masters
Lew Masters received his BSME
degree in 1966 from the University
of Maryland and his MSEE degree
in 1969 from the University of
California at Santa Barbara. An HP
employee since 1970, Lew has
designed several modules for the
5300A Counter Mainframe and
was project leader for the 5381 A/
82A/83A Frequency Counters and
the 531 5A Universal Counter. Be
fore coming to HP, he was in
aerospace mechanical design.
Raised in Washington, D.C.,
Lew is married and lives in Los
Altos, California. He enjoys building furniture and remodeling
his home in his spare time.
Time=16 hr
Fig. 6. Three-stage hysteresis charging cycle for the leadacid batteries.
of the comparator and the process is repeated. This
second stage of operation, with the battery charge
current switching between 500 mA and 10 mA, con
tinues until the battery voltage fails to reach the lower
threshold of the comparator in the 10-mA state. At
this point the battery is 95-97% charged. In the final
stage of the cycle, the battery is maintained at the 10
m A current level and the remaining 3-5% of charge is
returned to the battery. At full charge, the 10-mA
constant current causes a battery voltage of approxi
mately 2.4 volts/cell (7.2 volts). The battery can be
safely maintained at this level for standby applica
tions.
An additional benefit of the hysteresis charger is
that normal operation of the circuit provides useful
charge status information. An LED in the display sec
tion of the instrument monitors the activity of the
charge circuit comparator, indicating the three stages
of the charge cycle. During the first stage of the cycle,
the LED is off to indicate that less than 70% of the
discharged capacity has been returned and the
charger is in the high-current state. During the second
stage of operation, the LED flashes on and off to indi
cate the low and high-current states, respectively.
This stage occurs after approximately 70% of the pre
viously discharged capacity has been returned
(charge acceptance accounted for). During this stage,
the ratio of the time the charger spends in the highcurrent state to the time it spends in the low-current
state depends on the absolute state of charge of the
battery. This means that in addition to information on
the amount of charge returned, indicated by the flash
ing LED, the ratio of the LED's on time to its off time
gives an indication of the absolute state of charge.
In the third stage, the LED is on continuously, indi
cating that the battery is at least 95-97% charged and
the charger is in the low-current state.
A temperature compensation feature of the charger
circuit provides proper charging and protection for
Karl M. Blankenship
Karl Blankenship received his
BSEE degree from Stanford Uni
versity in 1973 and joined HP the
same year. Before taking on the
power system design for the
531 5A/B Universal Counter, he was
responsible for the design of sev
eral circuits and options for the
5328A Universal Counter. Born in
Los Angeles, Karl is married, has a
son, and lives in San Jose, Califor
nia. He enjoys bowling, basket
ball, and jazz, writes short stories
and poetry, and spends a lot of
time on family activities, especially
supporting his son's budding gymnastics talent.
Michael J. Ward
Mike Ward designed the input
amplifiers and associated circuits
for the 531 5A/B Universal Counter.
With HP since 1973, he was in
volved in the design of several
boards and products for HP's
laser interferometer systems be
fore switching to the counter lab.
Mike was born in Worcester, Mas
sachusetts. He received his BSEE
degree in 1966 and his MSEE in
1 968 from Massachusetts Institute
of Technology, and did RF and
microwave design for an aeroi space company before coming to
HP. He's a member of IEEE. In his spare time, he edits and
publishes a small magazine about science fiction. He's also a
fan of railway and transit systems and of old-time jazz. Mike is
single and lives in Sunnyvale, California.
10
© Copr. 1949-1998 Hewlett-Packard Co.
instrument is operating from the battery and the bat
tery voltage is above the auto shutdown point. Auto
shutdown is indicated when the same LED flashes at a
slow rate. This reduces the possibility of unintention
ally discharging the battery and informs the user
when recharge is necessary.
vides automatic shutdown and discharge status an
nunciation. Automatic shutdown is controlled by the
low-battery detector circuit. This circuit consists of a
voltage comparator with hysteresis that monitors the
battery voltage during discharge. When the voltage
falls below the low threshold of the comparator, the
change of state of the comparator output is sensed by
the dc-dc converter and regulator section of the main
power supply. As a result, the operation of these cir
cuits is disabled and the current drain on the battery is
reduced to approximately 10 mA. At this point, ap
proximately 80% of the battery's rated capacity has
been used, allowing for a reserve capacity of 20% . The
instrument can remain in this state for 4-5 days with
out damaging the battery.
An LED in the display section of the instrument and
circuitry that monitors the activity of the low-battery
detector provide useful annunciation during battery
operation. The LED is on continuously whenever the
Acknowledgments
The project team would like to thank a few of the
many individuals who helped make the 5315A/B pro
ject successful. Carl Spalding contributed the product
design with assistance from Bill Anson and Eric
Havstad. Jim Feagin contributed many ideas on how
to make the instrument testable and producible. Russ
Aleshi and Bill Ruhl provided the automatic test
programs, while Bob Boss and Jim Conforti guided
the instrument through pilot run. Finally, Jim Homer
gave us a clear goal and provided the support neces
sary to achieve it.ff
ABRIDGED SPECIFICATIONS
HP Model 531 4A Universal Counter
HP Model 5315A/B Universal Counter
INPUT CHARACTERISTICS: (Channel A and Channel B)
RANGE:
DC coupled: 0 to 100 MHz.
Ac coupled: 30 Hz to 100 MHz.
SENSITIVITY:
10 mV rms sine wave to 10 MHz.
25 rnV rms sine wave lo 100 MHz.
i pulse width of 5 ns. Sensitivity can b
75 mV peak-lo-peak pulse at i
NOMINAL by adjusting sensitivity cor
varied continuously up to 500
ity mode trigger level is automatically sel lo OV NOMINAL
I rol In ser
i r dC switchable
COUPLING
IMPEDANCE
1 Mlt NOMINAL shunted by less lhan 40 pF
500 Ml NOMINAL shunied by less than 70 pF (COMMON A).
ATTENUATOR, xt or *20 NOMINAL.
TRIGGER LEVEL Variable between +2V and 2V.
SLOPE: Independent selection of • or slope
CHANNEL INPUT. SEPARATE or COMMON A
FREQUENCY:
RANGE: 0.1 Hz to 100 MHz.
LSD* DISPLAYED. 10 Hz to 1 nHz depending upon gate t
input signal. Al least 7 digits displayed per second of gate time
PERIOD:
RANGE; 10 ns to 10 s.
Is depending upon gate lime and u
LSD' DISPLAYED. 100 ns
per second of gate lime
signal. At least 7 digits displ
TIME INTERVAL:
RANGE: 100 ns to 10s S
LSD' DISPLAYED: 100 ns
TIME INTERVAL AVERAGE:
RANGE 0 ns to 105 s.
LSD* DISPLAYED. 100 ns to 10 ps depending upon gale lime and mpul
3 start) 200 ns
TIME INTERVAL HOLDOFF (Delay): Front panel gate lime knob inserts a variable
delay of NOMINALLY 500 (is to 20 ms between START (Channel A) and enabling
of STOP (Channel B). Electrical inputs during delay time are ignored. Delay
time may be digitally measured by simultaneously pressing T I Averaging.
if T i. Holdofl s
Delay, and blue key Other specific
e Interval
RATIO:
RANGE: 0.1 Hz lo 100 N :. both channels
LSD:
2.5 x Period A
;t decade)
>unded tc
Gale Time
TOTALIZE:
MANUAL Range: 0 to 100 MHz
A GATED BY B Totalizes input A between two e1
reset to make new measurement Gate opens n A slope, closes on 6 slope
Range: Oto 100 MHz.
TIME BASE:
FREQUENCY: 10 MHz.
AGING HATE • 3 parts in 10'imo.
TEMPERATURE. - -5 parts m 10*. 0 to 50 C.
GENERAL:
CHECK. Counts iniernal 10 MHz reterence frequency over gate 1i ; range
NOMINALLY 500 ^s to 20 ms.
DISPLAY: 6-digit amber LED display, with engineering unrts annunciator.
GATE TIME Continuously variable. NOMINALLY from 50ms lo 10 sor 1 period
SAMPLE RATE. Up to 5 readings per second NOMINAL except in time mierval
mode, to it is continuously variable NOMINALLY from 250 ms to 10s via
Gate Time control
OPERATING TEMPERATURE: O'C to 50 C
POWER REQUIREMENTS 100. 115. 210. 230V {-i-5%, -10%) 48-66 Hz.
10 VA max
WEIGHT 2 9 kg (6 10 5 OZ)
DIMENSIONS. 238mm W > 98mm H - 276 mm L (WiO*»» 105» in).
OPTIONS:
OPTION 001 High Stability Time Base (TCXO)
FREQUENCY. 10MHz.
AGING RATE 1 part in 107;mo.
TEMPERATURE t part in 10", 0 to 40 C
OPTION 002 Battery
TYPE Rechargeable lead acid (sealed).
CAPACITY TYPICALLY 4 hours cl continuous operation al 25 C
RECHARGING TIME TYPICALLY 16 hourb to 98% of Ml charge, mstrumetil
nonoperating Charging circuitry included with Option Batteries not charged
during instrument operation
LOW VOLTAGE INDICATOR: Instrument turns itself off automatically when
low battery condition exists DISCHARGE LED flashes slowly when this
happens DISCHARGE LED is on whenever battery is supplying powef to
CHARGE is indÃ-cales slate of charge ot batiery during charging only and is
on whenever battery is charged to 95% NOMINAL of capacity
CHARGE LED Hashes when 90% NOMINAL of charge taken oui is replaced
Charge LED is off if charge is less than 70% NOMINAL ol capacity,
LINE FAILURE PROTECTION. InsUument automatically swiiches to battery
in case ol line failure.
WEIGHT. Option 002 adds 1.4 kg (3 Ibs ) lo weight ol mslrurnent
531 SB:
Rack as stack metal case, ac line power only Specifications same as 5315A
except as followsOSCILLATOR OUTPUT 10 MHz. 50 mV p-p into 5011 load.
EXTERNAL FREQUENCY STANDARD 10 MHz, W rms into 5001!
DIMENSIONS 215mm W . 81 mm H • 279 mm L {B^-3''t • I0''e in)
WEIGHT 4 kg <B Ib 13oz|.
PRICES IN USA
5315A 100 MHz'100 ns Universal Counter. J80O
531 58 100 MHz 100 ns Universal Counter in Metal Rack.Stack Package. $950
OPTIONS:
001 High Stability Time Base. $100
002 Battery (available with 5315A only), 5225
•LEAST SIGNIFICANT DIGIT (LSD) DISPLAYED (S315A<B):
F R E Q U E N C Y 7
£5-'.1- — . FREO FREO 10 MHz
All above calculations should be rounded tc
become 1 Hz and 0 4 ns will be 0 1 ns)
TIME INTERVAL AVERAGE.
100 ns
I t o 2 5 i n t e r v a l s
25 10 2500 intervals
2500 10 250.000 iniervais
250.000 to 25,000.000 intervals
25,000,000 iniervais . .
100 ps
10 ps
11
© Copr. 1949-1998 Hewlett-Packard Co.
INPUT CHARACTERISTICS:
RANGE Channel A 10 Hz to 100 MHz.
Channel B 10 Hz ¡o 2 5 MHz
SENSITIVITY:
Channel A 25 rnV rms lo 100 MHz
75 mV peak-to-peak minimum pulse width ol 5 ns
Channel B 25 mV rms to 2.5 MHz.
75 mV peak-to-peak minimum pulse width of 200 n
COUPLING ac
IMPEDANCE 1 Mil NOMINAL, shunted By less than 30 pF.
ATTENUATOR *1 or -20 NOMINAL (A Channel only)
TRIGGER LEVEL. Continuously variable ±350 mV limes attenuator
average value ol signal.
SLOPE Independent selection ol • or - slope.
CHANNEL INPUT. Selectable SEPARATE or COMMON A
FREQUENCY:
RANGE 10 Hi lo 10 MHz direc! count,
10 Hz lo 100 MHz prescaled by 10
LSD DISPLAYED. Direct count 0 1 Hz. 1 Hz, 10 Hi switcr
Prescaled 10 Hz. 100 Hz, 1 kHz switch selectable
PERIOD:
RANGE 10 Hz to 2.5 MHz
LSD DISPLAYED
'lorN= tlolOOOindecadtj steps oIN
TIME INTERVAL:
RANGE 250 ns to 1 S.
LSD DISPLAYED too ns.
RATIO (A 10 B):
RANGE 10 Hz lo 10 MHz Channel A.
10 Hz to 2.5 MHz Channel B.
LSD DISPLAYED. 1/N in decade sleps oi N for N-l lo 1000
TOTALIZE:
RANGE. 10 Hz to 10 MHz
Totalize controlled by front panel switch.
GENERAL.
CHECK. Counts internal 10 MHz oscillaiot
DISPLAY 7-<*gil amber LED display with gate and overflow indication
MAXIMUM SAMPLE RATE. 5 readings per second.
OPERATING TEMPERATURE: 0' lo 50'C.
POWER REQUIREMENT. 115, ' 10%, 25%, 230V 17%, t9%. 48-66 Hi
1 0 VA maximum.
WEIGHT 20kg (4 4 Ibs, >
TIME BASE'
FREQUENCY 10MHz.
AGING RATE 3 parts in 10' per month.
TEMPERATURE: ' 1 part in 10s. 0 to 50 C.
OPTIONS:
OPTION 001. High Stability Time Base (TCXO)
FREQUENCY 10 MHz.
AGING HATE 1 part in 10' per month
TEMPERATURE 1 pati in 10'. 0 to 40 C
OPTION 002 Ballery
TYPE Rechargeable lead-acid (sealed;.
CAPACITY. TYPICALLY 8 hours of continuous operation al 25'C.
RECHARGING TIME TYPICALLY 16 hours to 98% ol full charge, instrumenl
nonoperating Charging circuitry included with option Batteries not charged
during instrument operation .
BATTERY VOLTAGE SENSOR. Automatically shuls instrument off when low
battery condition exists
LINE FAILURE PROTECTION Instrumenl automatically switched to batteries
in case of line failure.
WEIGHT: Option 002 adds 1.5 kg (3 3 Ibs.) to weighl of instrument.
PRICES IN U.S.A.:
5314A 100 MHz.'100 ns Universal Counter. $375
OPTIONS
001 Hiuii Stability Time Base. $100
002 Battery. $95
MANUFACTURING DIVISION: SANTA CLARA DIVISION
5301 Stevens Creek Boulevard
Sania Clara, California 95050
A High-Performance Bipolar LSI Counter
Chip Using EFL and I2L circuits
This state-of-the-art multiple-register counter chip
contains all of the circuits needed for a 100-MHz universal
counter except for the display, input amplifiers, power
supply, and controller (microprocessor).
by Bosco W. Wong and William D. Jackson
included in the MRC. Instead, the processor performs
all necessary calculations, with measurement data
furnished by the MRC chip.
The MRC is designed not for a single counter but
rather for a number of different counter products with
a variety of performance requirements. Thus, its per
formance capability is broad and flexible. Depending
on the type of processor and/or accessory circuits
used, the MRC can provide a wide range of measure
ment capability, including different modes of fre
quency, period, time interval, ratio, totalizing, and
other measurements.
MICROPROCESSORS HAVE BROADENED per
formance capability in many new counter prod
ucts. Intricate computing abilities are now readily
obtainable in commercially available microprocessor
chip sets. However, microprocessors alone cannot
carry out complex counting functions. In addition to
the processor unit, various circuit elements, both
digital and analog, are required to accept, store, and
regenerate data for different phases and different
types of measuring functions.
As a group, these circuit elements cover a wide
range of performance requirements. To optimize their
performance, we felt that it was highly desirable if not
totally essential to implement them in a special inte
grated circuit chip. For this reason, a custom bipolar
LSI chip known as the multiple-register counter
(MRC) has been developed for the new generation of
microprocessor-controlled counter products. Except
for the controller (microprocessor), the display cir
cuitry, the signal input amplifiers, and the instrument
power supply, all of the circuits required to perform
all necessary counting functions are included in the
MRC chip.
The decision to implement the MRC chip in the
bipolar LSI process is mainly based on three factors.
First, the process can accommodate mixed logic
families with divergent performance characteristics,
namely emitter function logic (EFL) and integrated
injection logic (PL). Second, the process can accom
modate miscellaneous circuit configurations, both
digital and analog. Third, the dual-layer metalliza
tion feature of the process provides an efficient means
of solving the complex high-density logic intercon
nection problem.
Multiple-Register Structure
The name multiple-register counter is derived from
the chip's logic organization, a parallel multi-register
structure. Fig. 1 is the basic functional block diagram
of the MRC. The four registers — the E (event) , T (tim
ing reference), C (control) and S (status) registers —
operate in parallel for I/O access.
The E and T registers are chains of eight decades
each, with the front-end (first) decades capable of
counting signals over 100 MHz. Because of the divi
sion factor of 108 resulting from the eight-decade se
rial arrangement, the signal frequency coming out of
the tail-end (last) decades is only 1 Hz for a 100-MHz
input frequency. To optimize speed, power, and de
vice density, it is obvious that different types of cir
cuits should be used for different segments of the
chains. Accordingly, the first two decades of the E
and T registers are implemented as EFL circuits to
meet the speed requirements, while the remaining six
decades are PL circuits for high device packing den
sity and low power consumption.
Signal counting takes place in these two registers
when gated by the sequencer block. An overflow flipflop is provided for each of the two registers to extend
the effective length of the counting chains.
The 20-bit control register serves as a storage buffer
for program instructions received from the processor
through the I/O block. As the name implies, it controls
the general operation of the MRC chip. Specifically,
A Data-Acquisition Chip
The MRC chip has been specifically designed for
universal counter use. The chip performs all the
data-acquisition functions of a universal counter or
timer when controlled by a microprocessor. For in
strument performance flexibility and chip design
simplicity, data computation features have not been
12
© Copr. 1949-1998 Hewlett-Packard Co.
Key:
Interrupt
Logic
High-Speed Circuitry
Low-Speed Circuitry
Event Register— 8 Decades
Sequencer
(E
Register)
Channel 4-to-3
Input MultiBuffer plexer
cades
(T Register)
Reference
Frequency
Input
Buffer
Status Register— 12 Bits
(S Register)
Control Register — 20 Bits
(C Register)
Oscillator
(Optional)
Ã-
I/O Data Bus Address Decoder
Buffer
I Ã-
Ã-
Fig. 1. Multiple-register counter
(MRC) chip functional block dia
gram.
sequencer logic block is most critical to the complex
counting capability of the chip. In the sequencer
block, two key elements that enable the MRC to excel
in performance are a sophisticated gating synchroni
zation technique and the realization of this technique
with EFL and other circuits in a super-speed bipolar
LSI process.
The gating technique involves three stages of suc
cessive synchronization: arming synchronization,
synchronization for E-register gating, and synchroni
zation for T-register gating. This scheme provides
maximum measurement resolution in the sense that
measurement error is limited to ± 1 count of either the
input frequency or the reference oscillator frequency,
whichever is higher.
The bipolar LSI process features dual-layer metalli
zation, which reduces on-chip parasitics, and an fT of
1 GHz for fast device switching.
Complementing the sequencer logic block are the
high-speed three-channel input buffer and the high
speed 4-to-3 multiplexer. Signals to be measured are
received through the channel inputs. They are then
directed into the E and/or T counting registers via the
multiplexer and the sequencer. The multiplexer
block is responsible for proper signal routing based
on the selected measuring function, while the se
quencer logic block is responsible for gating syn
chronization.
the jobs of the C register are to define the particular
mode of the measuring functions, to set up proper
synchronization and timing of the sequencer block,
and to carry out the reset routine before a measure
ment is initiated.
Operational status of the MRC, including mea
surement cycle arming and gating, overflow of the
two counting (E and T) registers, measurement inter
rupt and time-out, and so on, is recorded in the 12-bit
status register and is available to the processor when
properly addressed.
To minimize package pin count, the four-bit I/O bus
is bidirectional (see Fig. 1). The bus is shared by the
four registers. Through the address decoder block,
any of the four registers can be selected. The C register
uses the bus for the data input function (i.e., the write
function), and the E, T and S registers use the bus for
the data output function (i.e., the read function). Once
a read or write function has been requested and a
particular register has been selected, individual de
cades (in the case of the E and T registers), or indi
vidual words (in the case of the C and S registers), are
accessed sequentially by scanning the 3-to-8 nibbleselect logic within the address decoder block.
Sophisticated Gating Functions
Of all of the functional blocks in the MRC, the
13
© Copr. 1949-1998 Hewlett-Packard Co.
Special care has been taken to minimize crosscoupling between adjacent high-speed input signals,
to provide adequate ground connections (or taps) to
minimize IR drop for the ground-potential-sensitive
PL circuit blocks (see box, page 17), to alleviate heatgradient problems resulting from uneven power con
sumption of different types of circuits in different
locations of the chip, to ensure adequate noise margin
by generating on-chip temperature-compensated re
ference voltages for the EFL circuits, and to provide
additional test probe pads to reduce the normally
lengthy time required for testing multidecade chains.
Fig. 2 is a photomicrograph of the MRC chip. To
show the difference in device density, the cell layouts
of an EFL decade and an PL decade are marked.
Assortment of Circuits
Aside from conventional logic elements, such as
multi-input logic gates, flip-flops, decades, and so on,
which are implemented extensively in EFL or PL
throughout the chip, a variety of other specialfunction circuits, both digital and analog, are used in
the MRC. The digital circuits include miscellaneous
types of input-output buffers, high-speed and lowspeed multiplexers, different types of delay circuits
for timing use, special purpose complementary clock
drivers, matrix decoders, EFL to PL level converters,
and vice versa. The analog circuits include precision
voltage regulators, Schmitt triggers, oscillators,
differential amplifiers, and voltage level detectors.
A few of the MRC circuits are unusual enough to
warrant further discussion. These include the timing
delay circuit, matrix decoder, and level converters.
Fig. 2. MRC chip layout. EFL circuits are used where high
speed is needed. I2L is used for low power consumption and
high density.
Circuit and Mask Design Requirements
The diverse circuit performance requirements, in
cluding wide deviations of device speed, variable
packing density, mixing of logic families, on-chip
interplay among various types of digital and analog
circuits, and narrow noise margins (because of the
low supply voltage used), have made implementation
of the MRC more than a trivial task. Aside from com
ing up with an adequate assortment of digital and
analog circuit elements to realize various circuit func
tions, considerable engineering effort has also been
given to planning and design of the mask topology.
I2L Delay Circuit
The timing delay circuit is implemented in lowcurrent PL gates. In the past, 1C designers have had
difficulty implementing well-controlled long-delay
-v
R (Can Be Replaced with Current
Source for Better Regulation.)
Scale Factor for Injector Current
of Individual I2L Gates.
Out
Delay Elements
(with Low Injector Current)
Booster Stage
(with Increasing Injector Current)
14
© Copr. 1949-1998 Hewlett-Packard Co.
Fig. 3. I2L delay circuit provides
delays in the millisecond range.
plemented in special serially connected PL gates.
provides delay times not only of microseconds but of
milliseconds. Fig. 3 shows the relatively simple cir
cuit. The delay is strictly a function of the number of
series PL gates used and the magnitude of the injector
current into each individual gate.
For long delay requirements, the injector current
should be held to a minimum for the bulk of the delay
string. However, for adequate drive to other circuit
blocks, the final section of the delay string should
have increased injector current, as shown in Fig. 3. If
the delay circuit is properly designed, the delay time
will be within ±20% of target value. Further accuracy
circuit elements with conventional bipolar circuit
techniques. The difficulty is mainly due to two fac
tors, namely the relatively fast transit time of conven
tional bipolar transistors and the impracticality of
making large-value on-chip capacitors and resistors.
It takes a moderately complex circuit and tight pro
cess control to generate a signal delay of up to a few
microseconds. Longer delays usually require off-chip
RC trimming.
By taking advantage of the intrinsically slower
propagation delay of PL transistors and by appro
priately controlling the magnitude of the injector cur
rent in the PL gates, the MRC delay circuit, im
Emitter Function Logic
by Bosco W. Wong
Reference
Emitter function logic (EFL), is a relatively new type of logic
circuit that is particularly useful for LSI implementation
whenever high switching speed is required. Developed by Z.E.
Skokan of Hewlett-Packard Laboratories, EFL is designed to
capitalize on two logic circuit features, namely the multi-emitter
structure normally found in TTL, and the nonsaturating current
mode operation used in ECL. To minimize propagation delay,
EFL uses noninverting gates. Its intrinsic gate delay is consider
ably less than that of other common logic families, such as TTL,
which use inverting gates.
EFL bears some degree of resemblance to the popular ECL
family. However, EFL exceeds ECL in cell density and in other
aspects of performance. While both EFL and ECL use the nonsaturated operating region of switching transistors for fast
switching performance, EFL requires a reduced total bias volt
age of only 3V, compared with the typical 5.2V of ECL. Also, an
efficient circuit partition technique known as functional integra
tion, employed exclusively in EFL, greatly reduces the overall
complexity of the circuit compared with ECL.
The most prominent feature of EFL is the functional integration
capability. Boolean functions can be directly implemented in
EFL gates without resorting to the conventional transformation
procedure. This means that Boolean equations need not be
converted into formatted structures like NAND and NOR expres
sions before circuit implementation. Essentially, each basic EFL
gate, with appropriate variations, can be used to realize differ
ent sets of multiple logic functions, such as sum-of-products or
product-of-sums expressions.
To illustrate the functional integration aspect of EFL gates,
Fig. 1 shows two EFL gate elements, each representing a differ
ent Boolean expression. The first gate (Fig. 1a) represents a
simple two-input AND function involving only one Boolean gate.
The second EFL gate (Fig. 1b) represents a product-of-sums
expression equivalent to four Boolean gates. A brief comparison
of the two EFL gate elements reveals that while the component
counts for both elements are almost the same (i.e., four transis
tors and two resistors in one element and five transistors and
one resistor in the other), the Boolean gate counts for the two
EFL gate elements are quite different (i.e., by a 1:4 ratio). It is
important that EFL designers be aware of this flexible capability
of EFL and take care to maximize their use of it.
1. Z. IEEE "Emitter-Function-Logic Logic Family for LSI," IEEE Journal of SolidStale Circuits, October 1973.
Out
Out=A-B
Single Boolean Gate
(a) EFL Cell to Realize Two-Input AND Boolean Function
Out
Out
Out=(A+C)-(B+C)
Four Boolean Gates
(b) EFL Cell to Realize Three-Input Product-of-Sums
Boolean Function
Acknowledgment
The author would like to thank Z.E. Skokan for his many
phases of technical support in EFL.
Fig. the . Two different EFL cells illustrate the flexibility of the
functional integration feature of EFL.
15
© Copr. 1949-1998 Hewlett-Packard Co.
mask-topology features, the dense packing layout of
the PL gates and the convenience of logic hook-up
using dual-layer metallization.
EFL-to-l2L Converter Circuits
Since PL gates operate at moderately slow speeds,
the interface between EFL and PL gates is reasonably
simple. Fig. 5 shows two different voltage conversion
schemes, one from EFL levels to PL and the other
from PL to EFL.
In the EFL to PL converter, the output transistor is a
multicollector inverse NPN transistor, that is, a typi
cal PL gate without an injector. Since switching
speed is not a prime requirement, lateral PNP transis
tors have been conveniently used in both converters.
-Me
a. Circuit Schematic
— 87 Se 85 84 83 82 81 So
Diffusion Contact for
Top Layer Metal Bus
Top Layer Metal Bus
.*- Bottom Layer Metal Bus
/-- Diffusion Contact for
Bottom Layer Metal Bus
System Interface
Since the MRC is designed for a number of different
counter products with a wide range of performance
specifications, interface flexibility has been em
phasized in the architectural design. The perfor
mance capability of the MRC depends largely on the
type of controller. A prescale option has been pro
vided for higher front-end counting frequencies and
extended resolution. All input-output pins, except for
the analog areas, are either TTL or ECL compatible.
87 SB 85 84 S3 S2 81 So
b. Mask Layout Diagram
Note: Ground Connections (to Epi) Not Shown
Top Layer
Metal
c. Cross-Section of Mask Layout
Bottom Layer
Metal
Vcc (3.0V)
(Buried Layer)
Fig. 4. 3-to-8 address decoder uses a double-injector
scheme, with injector rails flanking both ends of the long
inverse A/PA/ transistors. Note the dense packing of the I2L
gates in the convenience of dual-layer metallization for in
terconnections.
can be achieved by replacing the resistor R with a
well-regulated current source.
Matrix Decoder
Fig. 4 shows the 3-to-8 address decoder, another
circuit conveniently implemented in PL. However,
the design of this decoder differs slightly from the
conventional way of using PL (see Fig. 4b). Because of
the array layout, the multicollector inverse NPN
transistors are longer than those of regular PL gates.
To assure proper current drive for the worstpositioned collectors, which are the ones farthest
from the injector rail, a double-injector scheme has
been used, with injector rails flanking both ends of
the long inverse NPN transistors.
The mask layout sketch and cross-section for the
matrix decoder, Figs. 4b and 4c, also reveal two
(b)!2LtoEFL
Fig. 5. Voltage level converters used on the MRC chip.
16
© Copr. 1949-1998 Hewlett-Packard Co.
Integrated Injection Logic
by Bosco W. Wong
A youngster among bipolar LSI circuit families, integrated
injection logic (I2L) has been looked upon by some LSI design
ers and users as a challenge to MOS in terms of device density
and device performance. One fact that can hardly be denied is
that I2L fills a desperate need for better device packing capabil
ity in bipolar LSI design.
The I2L gate has a rather simple structure. As shown in Fig. 1 ,
it consists of a lateral PNP transistor as a current source and a
multicollector inverse NPN transistor as an inverter. The current
source performs the function of a pullup resistor, but in most
cases requires much less chip area. Also, the use of the PNP
transistor as a current source makes it possible to have reason
able control of the magnitude of the injected current.
In the typical chip layout, the two transistors are merged into
one nonisolated cell. Furthermore, it is possible to lay out a chip
so that no isolation is required among different I2L gates. Hence
device density can be very high. The need for isolation is
minimized by taking advantage of certain properties of inverse
NPN and lateral PNP transistors (see footnote).
The three attractive features of I2L are high packing density
(-200 to 300 gates per mm2), low power-speed product (typi
cally bipo- and on-chip compatibility with other popular bipo-
B, O
Bn O
-OC,
-O Cn
Fig. 1. Typical I2L gate.
lar LSI families and analog circuits. These have prompted the
use of I2L circuit technology in a variety of bipolar LSI designs in
the last few years.
The disadvantages of I2L fall into two areas. These are the
limited fan-out capability of I2L gates and a low tolerance for IR
drop along lengthy voltage lines or injector rails. In view of the
limitations, I2L is better suited for the implementation of lowpower serial logic, such as decade chains, serial registers, and
delay lines, than for random logic functions. For random logic,
I2L generally involves heavy parallel signal buses to connect
circuit elements located in different areas of the chip.
•In a normal vertical NPN transistor the collector is in the N-doperJ epitaxial layer grown on the
silicon P The base is a P diffusion in this layer, and the emitter is an N diffusion inside the P
area. same the collectors of different transistors are not normally at the same potential, isolation
is required between transistors. In an inverse NPN transistor the emitter is in the epitaxial layer,
making conventional isolation unnecessary, since emitters are often tied together in conventional circuit
design.
A lateral the transistor consists of two nearby P diffusions in the epitaxial layer, so that the
Reference
emitter, base, and collector areas are side by side instead of one on top of the other. In I2L, one
1 K Hart IEEE A Slob. "Integrated Injection Logic: A New Approach to LSI," IEEE
Journal of Solid State Circuits, October 1972.
of the transistors, diffusions, the injector rail, forms the emitters of all of the PNP transistors, so these
transistors need no isolation.
have made contributions to different phases of the
development project. The authors would like to share
the satisfaction derived from the successful MRC with
Mike Catalano, who as a member of the design team
for a period of time, was responsible for some of the
novel design techniques. Special thanks should also
go to Kay Bushey, Kazuko Kikuta, Percy Smith, Jay
Thomas, Barry Welsh, Jim Grace, Ed Hilton, Larry
Triplett and Dale Nieman.iZ?
Also, a chip enable-inhibit control input is available
for convenient system multiplexing.
Acknowledgments
The MRC chip is a product of a solid team effort
within HP's Santa Clara Division. Numerous people
both in the 1C department and in the engineering lab
Bosco W. Wong
I Bosco Wong received his BSEE
| degree in 1967 and his MSEE in
1973 from California State UniverI sity at San Jose. After four years
with a major semiconductor com
pany, mostly doing MOS RAM de
sign, he joined HP in 1971. He's
designed bipolar ICs for HP hand' held calculators and served as
project leader for I2L development
work and for the MRC chip design.
He's also co-authored a pair of
papers on I2L. Bosco was born in
China and grew up in Hong Kong.
He's married, has two sons, and
now lives in Sunnyvale, California. His hobbies are table tennis,
social dancing, and gourmet food.
William D. Jackson
( Bill Jackson received his BSEE
if degree from the University of
Arizona in 1 970 and his MSEE from
the University of Santa Clara in
1976. With HP since 1970, he's
helped design the 5328A Univer
sal Counter, the 10544A Oscil
lator, and the MRC chip. He's
named as inventor on a patent on
three-state trigger lights for count
ers. Born in Tuscon, Arizona, Bill is
married and lives in Cupertino,
California. He enjoys tennis,
music, and travel.
17
© Copr. 1949-1998 Hewlett-Packard Co.
A Synthesized Signal Source with Function
Generator Capabilities
This fully programmable signal source spans 13 decades
in frequency (1 fjHz to 21 MHz) with synthesizer accuracy,
an d p r o d u c e s s i ne wav es , s quare wave s, tria n g le s, a n d
ramps with synthesizer stability and swept-frequency
convenience.
by Dan D. Danielson and Stanley E. Froseth
TWO TYPES OF SIGNAL SOURCES are finding
increasingly wide use in today's world. One is
the frequency synthesizer. Because of its programmability, the synthesizer finds many uses in automatic
test equipment and, because of its inherent accuracy
and high resolution, it also finds use as a local oscilla
tor for receivers and as a precision tunable source for
exacting tests, such as measuring crystal responses.
The other type, the function generator, is valued for
its versatility, its capability for generating very low
frequencies, its variety of waveform outputs, and its
low cost. Among the applications of function gen
erators are the step-response testing of loudspeakers
and other electromechanical devices, filter transient
response testing, amplifier and detector linearity test
ing, logic testing, control and timing, and tests in
volving materials, geophysics, and biomedicine.
The versatility of a function generator has now been
combined with the accuracy, resolution, and programmability of a synthesizer in the HP Model 3325A
Synthesizer/Function Generator (Fig. 1). As a fre
quency synthesizer, this instrument has a resolution
of 11 digits and its sine-wave output covers an ex
ceptionally wide range of frequencies, from 1 /u,Hz to
21 MHz — over 13 decades. Unlike other synthesizers,
the phase of the output can be adjusted over a ±720°
range with respect to an internal reference. Also, it
can sweep the entire frequency range or any part of it
with a phase-continuous sweep (no transients be
tween frequency steps). In terms of synthesizer per
formance, it has low sine-wave distortion (-65 dBc)
at audio frequencies.
As a function generator, the Model 3325A is
noteworthy because of its frequency accuracy,
waveform precision (triangles and ramps with
linearities within ±0.05%), low-frequency capa
bility (to 10~6 Hz), and programmability. Output
amplitude is adjustable up to 10 volts into 50ÃÃ and
the output can be offset up to ±4. 5 Vdc. The offset can
also be used by itself to provide a dc level up to ±5 V.
Provisions are also included for phase modulating
any output function and for amplitude modulating
the sine-wave output. An optional output amplifier
increases the output amplitude and offset by a factor
of four (into 500Ã1) up to 1 MHz.
Several convenience features in the new instru
ment include front- and rear-panel sync outputs, a
rear-panel 20-60-MHz high frequency output pro-
999 999. 999 •
Ç) roms in l^iM
Ç C 1; ü
18
© Copr. 1949-1998 Hewlett-Packard Co.
Fig. 1 . Model 3325A Synthesizer/
Function Generator produces
waveforms with frequencies as
low as 1 fiHz and as high as 21
MHz for sine waves (to 1 1 MHz for
square waves and 1 1 kHz for
triangles and ramps) with synth
esizer accuracy. Output ampli
tude over the full frequency range
is 1 mV to 10V p-p into 50flor, with
an option, to 40V p-p into 500ÃI
over a frequency range of 1 fjiHz to
1 MHz. All necessary functions
can be controlled through the HP
interface bus as well as the front
panel.
HP-IB Input
Microprocessor
Fractional-N
Synthesizer
Phase
Modulation
Input
Reference
Output
Reference
Phase Lock Loop
External
Reference
Input
X-Drive
111
Z-Blank
Marker
Fig. Generator. operations diagram of the Model 3325A Synthesizer/Function Generator. All operations are
controlled by the microprocessor in response to commands entered through the front panel or by
way of the HP-IB interface.
fractional-N technique, generates an output fre
quency in a range of 30-51 MHz that is mixed with a
30-MHz crystal-controlled reference to produce an
output in a range of 0 to 21 MHz. The 30-MHz refer
ence is also divided down to 100 kHz to serve as the
reference for the synthesizer. The mixer output is
amplified and then used directly as the sine-wave
output or it is used to generate the square, triangle, or
ramp functions.
The output level is set by a programmable 10 dB/
step attenuator and amplifiers with programmable
gain. Analog levels within the instrument are con
trolled by a multiple output D-to-A converter, to be
described later. A peak comparator at the output of
the final amplifier is used during an automatic
routine to calibrate the output amplitude.
All operations are controlled by the microproces
sor, which enhances the versatility and "friendli
ness" of the instrument. Operating conditions are
entered by way of the front-panel keyboard or through
the HP-IB interface.
grammable from the front panel, and a 1-MHz refer
ence output. The instrument can be phase-locked to
any subharmonic of 10 MHz between 1 and 10 MHz
and a high-stability oven-controlled reference os
cillator is available as an option.
All the functions within this instrument are con
trolled by a microprocessor, a custom-made 8-bit pro
cessor using 16K bytes of read-only memory. This
enables a broadened control capability, such as the
ability to store and recall up to 10 sets of complete
instrument settings, the ability to "tune" the output
frequency by incrementing any display digit with
up-down step keys as well as by entering a desired
frequency with the numerical keypad, and extensive
error checking that insures that the selected signal
parameters are allowed for the functions chosen.
Also, the microprocessor enables the output
amplitude to be selected directly in the desired units
(VP-P> vrms. or dBm/50u). Extensive self-test
capabilities are also included.
The instrument is also fully programmable through
the HP interface bus. It is thus readily incorporated
into HP-IB-linked automatic test systems.
Fractional-N frequency Synthesis
Fractional-N frequency synthesis is a technique for
locking a voltage-controlled oscillator (VCO) to a frac
tional harmonic of a reference. That is, the oscillator
frequency is equal to the number N.F times the refer
ence frequency, where N and F are positive integers.
Instrument Organization
Fig. 2 diagrams how the Model 3325A is organized.
The synthesizer section, which uses the single-loop,
•The HP 488-1975 bus (HP-IB) is Hewlett-Packard's implementation of IEEE Standard 488-1975 and
ANSI Standard MC1.1.
19
© Copr. 1949-1998 Hewlett-Packard Co.
From
Microprocessor
Error Signal
Count-by-F
Section
Fig. 3. Simplified block diagram
of the fractional-N frequency
synthesizer used in Model 3325A.
The basic circuit is a conventional
count-by-N phase-locked I/CO.
The count-by-F section imple
ments the fractional division.
Analog
Phase
Interpolation
cycle. For example, if 0.33 were added to the ac
cumulator contents each reference cycle, starting
from 0 the adder would overflow in the fourth refer
ence cycle, leaving 0.32 in the accumulator, and then
overflow again after three more reference cycles. Av
eraged over many cycles, the VCO frequency would
be N. 33X100 kHz.
A salient feature of this technique is that the
number of digits synthesized is determined by the
number of digits in the F register, which is limited
only by hardware considerations. In the 3325A, the
frequency resolution is 15 digits, with 11 digits dis
played on the front panel.
In the Model 3325A, the VCO frequency in the syn
thesizer is N.FxlOO kHz where N is between 300 and
600 and F can be any integer 12 digits long.
A simplified block diagram of the fractional-N fre
quency synthesizer is shown in Fig. 3. The compo
nents of a basic count-by-N phase-lock loop are pre
sent. That is, a pulse is sent to the phase detector for
every N VCO pulses counted and the phase of this
pulse is compared to a 100-kHz reference to derive a
phase-error signal for controlling the VCO frequency.
Hence, the VCO frequency is N times the reference.
Added to these basic blocks is the count-by-O.F
section. The fractional part, F, of the frequency is
entered into the F register in BCD form and stored
there. Once during each cycle of the 100-kHz refer
ence, the contents of the F register are added to the
contents of the phase accumulator. Whenever this
addition causes the adder to overflow, one cycle of the
VCO output is deleted from the count-by-N input so
effectively, N+l VCO pulses are counted for that
particular reference cycle.
As an example, suppose that 0.1 is added to the
phase accumulator every reference cycle. The adder
then overflows every 10 reference cycles so in one out
of 10 reference cycles, N + l VCO pulses effectively
are counted. Thus, when averaged over 10 cycles, N.I
VCO pulses are counted per 100-kHz-reference cycle,
so the VCO frequency is N.I x 100 kHz.
Note that even when the adder overflows, its con
tents are transferred to the phase accumulator and so
are included in the total during the next reference
Phase Control
The basic count-by-N. F scheme just described re
quires certain refinements to become a viable
technique. For example, cycle removals cause the
phase detector to detect a 360° phase error every time
a pulse is deleted. To prevent the resulting large
transient in the phase-error signal from disturbing the
VCO loop, the analog phase interpolation (API) block
was added to the circuitry. It anticipates a pulse dele
tion and makes corrections to the integrator output so
large error signals are not transmitted to the VCO (see
Fig. 4).
A more detailed diagram of this part of the syn
thesizer is shown in Fig. 5. The first event that occurs
every reference cycle is that the VCO/N pulse from the
count-by-N circuit turns on the phase detector cur
rent, causing the integrator to ramp up. The next
20
© Copr. 1949-1998 Hewlett-Packard Co.
REF 23.8 ¿Bu
10 dB/DIV
DARKER S 000 010.0 Hz
RRNCE 29. O dB* 22.8 ¿Bu
REF 23.8 d
10 dB/DIV
CENTER 3 004 000.0 Hi 8PAM 10 000.0 Hi CENTER 9 004 000.0 Hz
RBU 100 Hi VBH 300 Hi 8T 2.0 SEC RBU 100 Hz VBU 300
100-kHz reference pulse turns it off. Thus, the level of
the integrator output after the ramp up is proportional
to the phase difference between VCO/N and the 100kHz reference. This level is retained by the sampleand-hold circuit and passed to the VCO. Following
the sampling, the bias signal turns on the bias current
which ramps the integrator down to the starting level.
When the desired output frequency is not an inte
gral multiple of the reference frequency, the VCO/N
pulse gains a fractional part of a cycle with respect to
the reference each time it occurs. Thus, until a pulse
deletion occurs, the phase detector pulse becomes
wider and the integrator ramps up further during
each succeeding reference cycle. It is therefore neces
sary to ramp down further each time so the integrator
ramp up will always end at the same level.
100-kHz
Reference
I 4'
Phase
Detector
Switch
NtfftMMflM*
Hz
SPAN 10 000-0 Hz
ST 2.0 SEC
Fig. 4. Spectrum displays show
the effect of analog phase interpo
lation (API). Spurious output fre
quencies are evident in the photo
at left, which shows the spectrum
of a 5.001-MHz sine-wave output
without the use of API. Photo at
right shows the spectrum of the
same output with API.
The necessary change in ramp-down current is con
trolled by the API switches which are in turn control
led by the phase accumulator. At the end of each
reference cycle, the number stored in the accumulator
corresponds to the difference in phase between the
VCO/N pulse and the reference. Each of the top five
decimal digits of this number controls one of the five
API bias switches, and turns on the switch for a period
inversely proportional to the numerical value of the
digit. The bias current is thus adjusted according to
the phase difference.
Since the number in the phase accumulator con
trols the phase of the VCO through the action of the
API currents, the VCO phase can be changed arbitrar
ily by changing this number. Hence, by adding an
increment to the F register for one reference cycle and
VCO/N
Sample-and-Hold Level
Sample-and-Hold Signal
Five Most
Significant
Digits
Phase Accumulator
Fig. 5. Details of analog phase
interpolation.
21
© Copr. 1949-1998 Hewlett-Packard Co.
then subtracting it in the next cycle, the output phase
is incremented with respect to the reference while the
frequency remains unchanged.
Phase modulation is accomplished by converting
the phase modulation signal into a current and sub
tracting this current from the bias current. Phase
modulation bandwidth, however, is limited by the
bandwidth of the phase-lock loop, which is about 7
kHz.
CH fl: UNCflLIBRflTED
10dB/DIV
Swept Frequencies
Phase-continuous frequency sweep of all functions
was made possible by the addition of another register
and BCD adder-subtractor to the count-by-O.F section.
A frequency increment stored in this register is added
to the F register during each reference period to cause
the frequency to sweep up. Sweep down is ac
complished by subtracting the increment.
The user has control of the start and stop frequen
cies, choice of linear or logarithmic sweep, and sweep
time. The linear sweep has single-sweep and con
tinuous-sweep modes with sweep times between 0.01
and 100 seconds. Total sweep times for single log
sweep range from 2 to 100 seconds or 0.1 to 100
seconds for continuous log sweep. X-axis and Z-blank
outputs are provided for driving CRTs and X-Y recor
ders. Also available is a TTL-level marker output that
changes state during a linear sweep when the output
frequency passes through the selected marker fre
quency.
Logarithmic sweep is approximated by a succes
sion of linear segments. A new frequency increment
is loaded 10 times per decade during single sweeps,
19 KHz '
BW: 60.0 Hz
Fig. 7. Spectrum of a typical sine-wave output at 1 kHz and
maximum amplitude (into 50Ã1J shows the relatively low distor
tion of Model 3325A at audio frequencies.
or twice per decade in repetitive sweeps (the sweep
width must be at least one decade). There is a short
pause in the sweep while the microprocessor com
putes the next frequency increment and as a result,
the X-axis output drive signal stops ramping up dur
ing the computation intervals. On a CRT display this
has the effect of placing brightened vertical bars,
which serve as convenient markers (Fig. 6).
Fractional-N 1C
The cost of the fractional-N digital circuits was
reduced significantly by designing them into a cus
tom NMOS integrated circuit. This 1C includes the
phase accumulator and F registers, timing for the
analog circuits, and the adder-subtractor and aux
iliary registers for implementing the sweep, but not
the high-speed divide-by-N circuitry. It also has a
sweep-limit register that signals when a preset fre
quency is reached during a sweep. This register is
used to generate the TTL marker during a sweep or to
stop the sweep at a preset limit.
The sweep-frequency register and addersubtractor, when not being used for the sweep func
tion, are used heavily by the microprocessor during
the math routines such as those that derive frequen
cies and amplitude levels. The microprocessor itself,
which is optimized for high-speed control functions,
has limited computation capability.
Waveform Generation
As shown in the block diagram of Fig. 2, the syn
thesized VCO output is sent to a signal-routing and
buffering block. This supplies a 20-to-60 MHz output
at a rear-panel connector (frequencies between 20 and
Fig. 6. Logarithmic sweep generates brightened vertical bars
as markers (two per decade in the continuous-sweep mode).
This oscil/ogram shows the output of a 500-kHz low-pass filter
when the input is swept logarithmically from 5 kHz to 5 MHz.
22
© Copr. 1949-1998 Hewlett-Packard Co.
product of the two input waveforms: a rectangular
pulse train with a pulse width that periodically
widens to 100% and narrows to 0%. A linear-phase,
low-pass filter removes the pulse transitions leaving
the dc component, a triangular wave.
Looking at this process in the frequency domain,
the gate output is the convolution of the spectra of the
two square waves, which is the spectrum of a triangle
wave with the difference-frequency components cen
tered at 0 Hz and the sum-frequency components
centered at 2 MHz. The low-pass filter removes the
sum-frequency components.
The reason for generating the triangles this way is
the exceptional linearity achieved, better than 0.05%
between the 10 and 90% amplitude levels (Fig. 9).
This is highly important when using the instrument
to check amplifier and A-to-D converter linearity.
However the maximum triangle frequency is limited
to 11 kHz. If higher frequencies were attempted,
the 1-MHz "carrier" frequency would have to be
raised and the pulse-train transition times would
then become an appreciable part of the pulse wave
form, degrading the resulting triangle linearity.
Ramp waveforms are generated similarly except
that when the ramp reaches its maximum level, a reset
pulse is generated. This inverts the 1-MHz reference
square wave, effectively advancing the phase of the
reference 180° so the ramp resets to the starting level.
Since two ramps occur within the time frame of one
triangle, the frequency added to the 1-MHz variable
input is one-half the desired output, e.g., for a 10-kHz
ramp repetition rate, the variable input would be
1.005 MHz.
Two methods are used to derive the reset pulse for
30 MHz are obtained by running the VCO in the 40to-60-MHz range and dividing its output by two).
The 0-20-MHz sine-wave output is produced by
mixing the VCO output with a 30-MHz reference. To
keep spurious outputs from the mixer at a low-level,
the VCO output drives a monolithic 1-GHz transistor
array to derive a fast-switching square wave for use as
the local oscillator (L.O.) signal in the mixer. The
mixer diodes are in a ring configuration using
Schottky diodes in a monolithic quad to achieve good
temperature tracking and balance. The typical 2:1 and
3:2 spurious products in the 0-20-MHz range are more
than 80-dB below the output signal. Harmonic distor
tion in the audio range typically is —70 dBc at full
output (Fig. 7).
Square waves are generated by using the 0-20-MHz
sine wave to drive a Schmitt trigger. This obtains a
fast-rise pulse train that is used to drive a -^2 flip-flop
that generates the square wave. This arrangement in
sures that the square wave is symmetrical. Since
it divides the sine-wave frequency by two, the
maximum square-wave repetition rate is 11 MHz.
Square-wave rise time is 20 ns and the use of de
coupled output circuits ensures fast settling time
(<l/xs to settle within 0.05% of the final p-p
amplitude).
Triangles and ramps are generated by using an
exclusive-OR mixing gate and waveforms with offset
frequencies. One input to the gate is a fixed, 1-MHz
reference square wave, as shown in Fig. 8. The other
input is a square wave with a repetition rate 1 MHz
higher than the desired output frequency, e.g., for a
10-kHz output this input would be 1.01 MHz. The
output of the gate, viewed in the time domain, is the
20 (1 MHz + Signal
Frequency*
Square Wave
Triange <
1 MHz + Signal Frequency" Spectrum at
^
D C
a n d
2
M H z
Triangle or Ramp
, Spectrum at
DC Only
Ramp Polarity Select
High=-Ramp
Low=+Ramp
1 MHz Reference
Ramp
Peak Voltage
Level from
D-to-A Converter
C Level
omparator
U 1¿is Ramp Reset Pulse
(Signal-Frequency 100 Hz)
One-Shot
Multivibrator
10Ms
Selects Reset Pulses
from Phase Detector
or One-Shot
] [•\0fis Ramp Reset Pulses
(Signal-Frequency 100 Hz)
23
© Copr. 1949-1998 Hewlett-Packard Co.
*Vi Signal
Frequency
for Ramps
Fig. 8. Circuits that generate the
triangle and ramp functions use
offset frequencies to obtain high
linearity in the output waveform.
reaches the peak value is used for generating the reset
pulse.
Amplitude and Offset Control
Vernier control of the output amplitude, control of
the dc offset, calibration of the amplitude and offset,
and control of the X-drive output signal are all ef
fected through a timeshared digital-to-analog conver
ter (DAC). A block diagram is shown in Fig. 10.
The heart of the DAC is the integrator. It ramps up
from a known voltage level at a rate fixed by the
precision current source for a period of time deter
mined by the preloadable countdown counter. The
clock frequency and the number loaded into the
counter thus determine what the integrator voltage
level will be at the end of the ramp.
The DAC continuously produces six output volt
ages. The sequence of events for producing any one of
them is as follows.
1. The microprocessor controller loads a four-digit
BCD number, N, corresponding to the desired out
put voltage, into the counter.
2. The integrator is reset.
3. The counter and the precision current source are
enabled simultaneously.
4. When the counter reaches 0 after counting down N
clock pulses, the current source is disabled and the
integrator holds at a voltage directly proportional
Fig. 9. Typical linearity error of the 3325A triangle wave out
put. This measurement was made on the positive slope ol a
10-kHz, 10-Vp.p triangle wave.
the ramps. For output repetition rates below 100 Hz, a
digital phase detector senses the coincidence of the
positive edges of the two square waves feeding the
mixing gate. Phase coincidence occurs at the peak of
the ramp. For very low frequencies, this gives a very
precise indication of the ramp peak, the resolving
capability of the phase detector being 1 fj.s, or one
reference period. However, at output repetition rates
above a few hundred hertz, this I-/K.S uncertainty is
manifested in an undesirable jitter. Thus, for outputs
above 100 Hz, a comparator that fires when the ramp
Sample-and-Hold
Amplifiers
Level
T~ Comparator
Reference
Start
(From Decoder)
Fig. 10. Time-shared digital-toanalog converter produces six
dc voltages.
24
© Copr. 1949-1998 Hewlett-Packard Co.
to N.
5. The integrator is sampled by the appropriate
sample-and-hold amplifier.
The cycle then repeats for the next value. Each
cycle requires approximately 1 ms so that each of the
six sample-and-hold outputs is updated every 6 ms.
The amplitude and offset control voltages are used
to set output waveform amplitudes and offsets. The
X-drive control voltage is used as the input voltage to
the X-drive integrator. The autozero output is used for
zeroing the DAC output. During the autozero cycle,
the number that is supposed to generate a zero-level
output is loaded into the counter. If the actual output
is different from zero, the voltage adjusts the current
source in a direction that tends to bring it to zero.
0-20 MHz Sine Wave
_
A
m
p l i
Signal
Current
f
i
Output
e r
To Output
Attenuator
son
Amplitude
Control
4-Digit DAC
and
S/H Circuits
DC Offset
Control
Level Reference
Level
Comparator
Amplitude and Offset Calibration
A routine for calibrating the output amplitude and
offset levels is executed whenever the instrument is
turned on, whenever the output function is changed,
whenever the AMPTD GAL button is pressed, and
whenever an external controller transmits a device
clear or an amplitude calibrate command.
Referring to Fig. 11, the amplitude and offset calib
ration routine consists of the following steps.
1. The controller attempts to set the signal amplitude
and offset to zero.
2. The controller finds the positive and negative peak
levels of the resulting output waveform by incre
menting the level reference voltage and using the
level comparator to indicate when a match is
achieved. The average value of the two peaks is the
dc offset correction and is subtracted from all sub
sequent dc offset settings.
3. The controller next attempts to set the signal
amplitude to 8 V p-p at 2 kHz.
4. The comparator and level reference voltage ors
again used to find the positive and negative peaks
of the output signal. The ratio of 8V to the mea
sured peak-to-peak value is used to scale all sub
sequent amplitude settings.
A complete calibration takes about 1.5 seconds and
assures offset and amplitude accuracy as good as 1%
(see specifications, next page, for details).
Fig. 1 1 . Circuits involved in the amplitude and offset calibra
tion routine.
As a convenience feature, the instrument's HP-IB
address is displayed when the BUS ADRS button is
pressed.
The HP-IB hardware is optically coupled to the
function-generating hardware to provide groundloop isolation.
Mechanical Construction
Unusual in a synthesizer, the 3325A was designed
with an open layout that makes the instrument easy to
assemble and service. This also allows the compo
nents to remain cool, thereby improving reliability.
The only shielding required is for sensitive analog
circuits in the VCO and the sine wave mixer. This is
done with low cost extrusions that attach to the main
circuit board.
Servicing was also simplified by designing the con
troller for digital signature analysis.1
Acknowledgments
Doug Garde, who was the original project manager,
conceived the instrument and designed the syn
thesizer. Rodger Kahnna and Ron Riedel designed the
fractional-N 1C. Ron also designed the function cir
cuits. Art Dumont designed the controller hardware
and Mike Price did the software. Dennis Faerber
handled the mechanical design. Much helpful advice
was provided by Ray Hanson and Noel Pace. R
HP-IB
All of the operations of the 3325A may be con
trolled through the HP-IB. There are two pro
gramming modes. In one, the 3325A processes each
ASCII character as it is received. Although this mode
provides ease of use, it can delay the HP-IB system by
causing it to wait while the 3325A processes each
character. In the other mode, the 3325A can accept
and store a string of up to 48 characters and not
attempt to process them until a string terminator is
received. The HP-IB controller may thus turn to other
tasks while the 3325A processes the character string.
Reference
1. R.A. Frohwerk, "Signature Analysis: A New Digital
Field Service Method," Hewlett-Packard Journal, May
1977.
25
© Copr. 1949-1998 Hewlett-Packard Co.
SPECIFICATIONS
OVERSHOOT: ^5% of peak to peak amplitude, al full output.
SETTLING TIME: <1,uS to settle to within .05% of final value, tested at full
output with no load.
SYMMETRY ERROR: -^0.02% of period - 3 ns.
TRIANGLE/RAMP LINEARITY (10% to 90%. 10 kHz): ±0.05% of full
outpu! lor each range
RAMP RETRACE TIME: 3 ps. 90% to 10%.
HP Model 3325A Synthesizer/Function Generator
WAVEFORMS: Sine, square, triangle, negative and positive ramps.
FREQUENCY
flANGE:
Sine: 1 ¿iHz to 20.999 999 999 MHz.
Square: 1 ,/Hz to 10999 999 999 MHz.
Triangle/Ramps. 1 ^Hz to 10.999 999 999 kHz.
RESOLUTION: 1 ^Hz. •; 100 kHz.
1 mHz 5*100 kHz.
ACCURACY. ±5.10 6 of selected value. 2(f 10 SO'C.
STABILITY: = 5 • 10 '6/yeai, 20° to 30°C.
WARM-UP TIME: 20 minutes to within specified accuracy.
REFERENCE INPUT: For phase-locking 3325A to an external frequt
reference. Reference signal, which must be a subharmomc of 1 0 MHz from
to 10 MHz. can be from 0 dBm to +20 dSm into SOfl.
MAIN SIGNAL OUTPUT (all waveforms)
IMPEDANCE: 5011.
CONNECTOR: BNC. switchable lo front or rear panel.
FLOATING Output may be floated up to 42V peak (ac-t-dc).
AMPLITUDE (all waveforms)
RANGE: 1 mV lo 10V p-p mlo SOU load in 6 amplitude ranges. 1-3-10 sequi
(10-dB stepsj.
DC Offset
RANGE:
DC ONLY (NO AC SIGNAL): 0 lo ±5.0 V/5011.
DC + AC: mV dc offset ±4.5 V on highest range: decreasing lo ±4.5 mV
on lowest range
RESOLUTION: 4 digits
ACCURACY:
DC only: ±0.006 mV to ±20 rnV depending on offset chosen.
Dc + AC: ±0.06 mV to ±60 mV to 1 MHz. ±15 mV to ±150 mV above 1 MHz.
depending on ac output level: ±0.02 mV to ±120 mV for ramps to 10 kHz.
OUTPUT IMPEDANCE: 500.
CONNECTOR Rear-panel BNC.
SYNC OUTPUT: Square wave with Vh(gn >1 2V, Vlow <0.2V ¡nio 5QÃI.
FREQUENCY RANGE: Same as main signal output.
OUTPUT IMPEDANCE; SOfl.
CONNECTOR BNC front and rear panels.
X-AXIS DRIVE: O lo -10V dc linear ramp proportional to sweep frequency.
0.1% linearity. 10-90%. Rear-panel 6NC connector.
SWEEP MARKER OUTPUT (linear sweep only): High to low TTL-compatible
voltage transition at key board- selected marker frequency. Rear-panel BNC
Z-AXIS current OUTPUT: TTL-compatible vot'age levels capable of sinking current
from a positive source. Currenl 200 m A. voltage 45V. power dissipation 1 watt
1-MHZ REFERENCE OUTPUT: 0-dBm output for phase-locking additional instru
ments lo 3325A. Rear-panel BNC connector.
10-MHZ REFERENCE OUTPUT: 0-dBm oulpul for phase-locking 3325A to trie
internal high stability frequency reference (Opt. 001). Rear-panel BNC
connector.
Phase Offset
HP-IB Control
RANGE: phase 7199'- with respect to arbitrary starting phase, or assigned zero phase
R E S O LU T I O N : 01=.
INCREMENT ACCURACY: -0.2"
FREQUENCY SWITCHING TIME (lo within 1 Hz. exclusive Of programming time):
s 10ms fot 100-kHz Step*; s25 ms fot 1-MHz step: «70 ms for 20-MHz step.
PHASE SWITCHING TIME: «15 ms to withtn 90' of phase lock, exclusive of
Sinewave Amplitude Modulation
AMPLITUDE SWITCHING AND SETTLING TIME: <30 ms to wilhm amplitude
MODULATION DEPTH (at full output for each range): 0-100%.
MODULATION FREQUENCY RANGE: Dc -50 kHz (0-21 MHz carrier frequency)
ENVELOPE DISTORTION: -30 dB lo 80% modulation at 1 kHz, 0 VDC offset
SENSITIVITY: ±5V peak for 100% modulation.
INPUT IMPEDANCE: 20 kit
CONNECTOR: Rear-panei BNC
Option 001 High Stability Frequency Reference
AGING RATE: ±5 • 10~8.'week. 1 <!0"7mo
ACCURACY: -5 - 10 "S ¡0; to - 50 Cl
WARM-UP TIME: Reference will be within ±1x10~7 of final value 15 minutes
after turn-on at 25rC lor an off lime of less than 24 hours.
Phase Modulation
RESOLUTION: 0 03% of lull range or 0.01 dB (4 digits).
ACCURACY is the sum of range and attenuator accuracies.
Range Accuracy (percent ot each range's maximum p-p outpu
Sine and Squarewaves:±l%. 10 Hz to 100 kHz.
±4%. 100 kHz to 20 MHz (sinewave
Triangles: ±1%. 10 Hz to 2 kHz.
±5%. 2 kHz to 10 kHz.
Ramps: ±1%, 10 Hz lo 2 kHz.
±10%, 2 kHz to 10 kHz.
Attenuator Accuracy (percent of output reading displayed;
when on 3- 10V range):
10 Hz to 10 kHz: ±1%. 1 m V to 2.999V.
10 kHz lo 10 MHz: -2%. 1 mV to 2.999V.
10 MHz to 20 MHz: ±2%. 0.1V lo 2999V.
±5%. 1 mV to 99.99 mV.
SINEWAVE SPECTRAL PURITY
20-MHz carrier (exPHASE NOISE: -54 dB for a 30-kHz band centered o
eluding ±1 Hz about the carr
SPURIOUS: All n vharrt
lutput
below tf
or less tÃ-
Option 002 High Voltage Output
SINEWAVE RANGE: ±850'. • 5V input
SINEWAVE LINEARITY: ±0.5' , Best fit straight line
SQUAREWAVE RANGE: -425
TRIANGLE RANGE: -42.5
POSITIVE AND NEGATIVE RAMPS: -85'.
MODULATION FREQUENCY RANGE: Dc-5 kHz.
INPUT IMPEDANCE: 20 - '
CONNECTOR: Rear panel BNC
Frequency Sweep
SWEEP TIME:
LINEAR. 0.01 s to 99.99 S
LOGARITHMIC: 2 s to 99.99 s single, 0.1 s lo 99.99 s continuous.
MAXIMUM SWEEP WIDTH: Full frequency range of the mam signal output for the
waveform in use.
MINIMUM SWEEP WIDTH:
FREQUENCY RANGE: 1 fMz to 1 MHz.
AMPLITUDE:
RANGE. 4.000 mV p-p to 40.00 V p-p (SOOÃÃ, <500 pF load)
ACCURACY: -2% of fuil output for each range at 2 kHz
FLATNESS: -10%.
SINEWAVE DISTORTION: Harmonically related signals will oe less than the
following levels (relative lo the fundamental full output inlo ï
10 Hz - 50 kHz' -65 dB
50 kHz - 200 kHz: -60 dB
200 kHz - 1 MHz: -40 dB
SOU ARE-WAVE RISE/FALL TIME: 100ns. 10% to 90% at ful
500-pF load.
i| pea k -to- peak ampliiude v
SQUARE-WAVE OVERSHOOT:
500-pF load.
OUTPUT IMPEDANCE: <2U at dc, • 1011 at 1 MHz.
DC OFFSET:
ACCURACY' ±(1% of full output for each range + ;
MAXIMUM OUTPUT CURRENT: 40 rtiA p-p
General
Waveform Characteristics
SINEWAVE HARMONIC DISTORTION: Harmonically related signals will be less
than full following levels (relative to the fundamental) at full output for each range:
1 0 H z
5 0 k H z
2 0 0 k H
2 M H z
1 5 M H z
t o 5 0
l o 2 0 0
z l o 2
t o 1 5
t o 2 0
k H z
k H z
M H z
M H z
M H z
- 6
- 6
- 4
- 3
- 2
SQUAREWAVE CHARACTERISTICS:
RISE'FALL-TIME. •- 20 ns 10% to 90%, at full output.
5
0
0
0
5
d
d
d
d
d
B
B
B
B
B
PHASE CONTINUITY: Sweep is phase contir
r the full frequency range
of the main output.
Auxiliary Outputs
AUXILIARY FREQUENCY OUTPUT:
FREQUENCY RANGE 21 MHz to 60.999 999 999 MHz.
to 19.000 000 001 MHz. frequency selection from fron
AMPLITUDE: 0 dBm.
Stanley E. Froseth
An HP employee since 1976, Stan
Froseth was responsible for por
tions of the analog design and
high voltage output option design
" of the 3325A Synthesizer/Function
Generator. He was born in St. Paul,
Minnesota, .and is a 1974 BSEE
and 1976 MSEE graduate of the
I
University of Minnesota. Stan is
also the author of several papers
related to environmental noise,
one of which was published in the
| Journal of Sound and Vibration. A
à ' x. Ã- member of the Loveland Environ^, ,PM» '<-^. -j^ mentai Quality Commission, Stan
also spends much of his spare time bicycling, playing tennis,
crosscountry skiing, and backpacking. Residents of Loveland,
Colorado, Stan and his wife have no children, but enjoy life at
home with their dog and seven cats.
OPERATING ENVIRONMENT:
TEMPERATURE: 0"C to 55 C.
RELATIVE HUMIDITY: 95%. 0°C to 40°C.
ALTITUDE: - 1 5,000 ft.
POWER: 100/120/220/240V. +5%. - 10%: 48 10 66 h z: 60 VA, 100 VA with a
options. 10 VA standby.
WEIGHT: 9 kg (20 Ib)
DIMENSIONS: 132 6 mm H x 425.5 mm W x 497.8 mn
. $550 High Vollage
PRICE IN U.S.A.: $3000. High -stability frequency refe
Output. $200.
MANUFACTURING DIVISION: LOVELAND INSTRUMENT DIVISION
815 Fourteenth Street. S.W.
Loveland, Colorado, 80537 U.SA.
Ã-
26
© Copr. 1949-1998 Hewlett-Packard Co.
Dan D. Danielson
Born in Grand Junction, Colorado,
Dan Danielson joined HP's Loveland Instrument Division in 1972,
shortly after receiving his BSEE
degree from the University of Col
orado. He began work as designer
ofthe output amplifier and modula
tion generator for the 331 2A Func
tion Generator and then worked on
the design of the synthesizing,
mixer, and reference circuits for
the 3325A Synthesizer/Function
Generator. A resident of Loveland,
Colorado, Dan is single and enjoys
skiing, golfing, and backpacking
in his leisure time.
Viewpoints
Paul Baird on Electronic Equipment Reliability
proved technology may increase the design margins and it may
also reduce the number of devices, resulting in lower X's. Error
correction involving redundancy, though effective, is practical at
present only in very complex digital products. However, reducing
temperature and electrical stresses can be particularly effective
countermeasures because X is very sensitive to these stresses.
Through application of these relationships, the designer can have a
powerful influence on the resulting product reliability.
Many of our modern electronic products are feasible because
advances in reliability accompanied those of technology and in
vention. Computers, for example, made relatively slow progress
until semiconductors replaced vacuum tubes. Then prices came
down while reliability improved greatly.
Reductions in price for a given performance have tended to mask
the improvements in reliability that are taking place. The reason is
that we commonly look at the cost of maintaining a product as a
percentage of its original price. Thus the manufacturer is kept
under constant pressure to meet customers' expectations with re
gard to those Actually, this is healthy since, by meeting those
expectations, powerful new products can win acceptance.
Materials Management
There is a residual set of potential failures that are best controlled
by device selection. Vendors frequently offer more than one level of
quality so some control is possible by this means. Then too, the
quality of a device varies from vendor to vendor and sometimes
from vendor, same vendor at different times. Thus, qualifying a vendor,
such of destructive stress testing of device samples and use of
incoming shipment acceptance procedures through some form of
inspection, are quite important. Vendors also do better as they gain
experience with their own products.
Experience has shown that the failure rates of most devices
improve with age, even over a span of several years. Thus, another
way to improve X is to accelerate the normal aging process, usually
referred to as burn-in.
In Search of Reliability
Reliability can be defined as the probability that a randomly
selected product will successfully complete an assigned mission. It
depends upon product robustness, mission stresses, mission dura
tion, and product age (history).
It is fairly common practice among those concerned with elec
tronic equipment to ignore the effects of product aging and use
mean-time-between-failures (MTBF) as the measure of reliability.
With such a model the reliability of a repaired product is no better
or worse than that of one that has never failed.
This isn't quite true — most products become more reliable with
age, at least for several years.
Also, because of the applicable mathematical relationships, most
engineers prefer to key their discussions on failure rate, denoted X,
rather than upon MTBF. The probability that a product will fail in
a time AT is approximated by XáT so that the smaller X is, the
more products the product is. Additionally, for nearly all products
(excepting those where redundancy techniques can be justified),
^product = •£ ^devices' These relationships hold whether or not X's
are time dependent, so we have left open our option to use sophisti
cated models for X if we choose to do so. Additionally, if the prod
uct \ does not change much with aging we can state MTBF =
Manufacturing
Following the selection of materials, some control over product X
is exerted by the manufacturing process. The manufacturing task is
to avoid workmanship flaws. Some of these may be latent, allowing
the product to work long enough to pass final inspection. Workers
basically want to do a good job so enlistment of their active support
through attention to their needs has a positive effect. Proper train
ing is results, important. Motivation and training produce best results,
however, when proper tools are available. As an example, many HP
divisions have electronic tools that check the tolerance of each
component after it is loaded on a printed-circuit board to assure
that it is still within tolerance following the mechanical stresses of
insertion and the heat stresses of soldering.
^product-
Since it is normally true that Xprodu(.t = 2 X(levk:es tne use °f
more devices means a higher failure rate unless certain countermeasures can be implemented. These countermeasures are ex
tremely important and are the subject of the following discussion.
After-Sale Support
Service support is a crucial item in how users feel about product
reliability. If repair turnaround times are short and the cost per
repair is low, customers tend to feel as if X were lower than it really
is. Since turnaround time and cost often conflict, service strategies
are often devised according to the product type. Computers, for
example, are usually serviced on-site because of the economic
importance of very fast turnaround time in this area.
Customer training can reduce "cockpit" errors that lead to
downed instruments. Good operating and service manuals also
have a of influence here. In addition, implementation of
proper preventive maintenance can reduce some types of hardware
failures.
In the Design Phase
Although nearly all field failures appear to be caused by a device,
most below design- or manufacturing-controllable. The graphs below
illustrate how product failure rate can be expected to vary as a
function of certain design-controllable factors. As the graphs show,
large design margins result in smaller X's since the product will not
be very are to drift in device parameters (design margins are
easily checked by non-destructive perturbation of internal vari
ables, such as supply voltages, noise, clock rates, etc.). An im
Device Count •
Design Margins-
Improved — i
Technology
27
© Copr. 1949-1998 Hewlett-Packard Co.
Error Correction/ — i
Tolerance Features
Stresses
*.
Temperature, Voltage
Power, Clock Rate,
Humidity, Vibration, .
Active Product Assurance
Results
Reliability is only partly technology-limited. It is also likely to be
limited by cost tradeoffs and by attention to detail in competititon
with other work objectives, so it is highly influenced by manage
ment commitment to reliability.
To achieve low X in their products, many companies are willing
to commit the resources needed to establish product assurance
departments. At Hewlett-Packard, product assurance managers,
most of whom have had broad experience in other departments,
provide the following services:
To act as a participant or else catalyst in the tasks already named.
Act as a conscience.
Provide information relative to quality and reliability.
Assure that somehow the proper things are happening.
Do quality and reliability engineering.
Assure compliance with the product safety and electromagnetic
compatibility regulations of the various nations where HP pro
ducts are sold.
Some of the major assurance techniques involved are listed here:
Looking ahead we expect to benefit from both statistical and
physical approaches to reliability. Statistics help set priorities by
quantifying problems and relationships. At HP, very complete
records are kept of warranty information about each serialized
product, detailing such items as ship dates, fail dates, repair office,
labor hours, turnaround time, a list of parts replaced (with failure
codes), total cost of repair, and free-form comments by the cus
tomer engineer.
Very useful analyses can be derived from this data base to pin
point troublesome models, components, or geographical areas.
Data is even complete enough to fit time-dependent Weibull failure
models to both products and components, which is done regularly.
After statistical definition of top problems we can concentrate
upon of the fundamental relationships in terms of
physics, chemistry, manufacturing processes and design stresses.
Most of HP's major divisions are devoting resources to high-stress
experiments on components followed by failure analysis of devices
that don't pass. Electromigration, for example, is a failure
mechanism now pretty well understood and controlled. On a selec
tive basis, failure analysis is also done on devices failing in en
vironmental tests, life tests, the factory floor, or in field use. Where
practical, results are shared on a company-wide basis. We have, for
instance, an ALERT system that can be activated by any .division to
warn the others of a possible problem. The ALERT also contains a
recommendation for dealing with the situation. As a result, some
problems are disappearing.
Reliability excellence in a product results from care in the design
of the product, in the choice of materials, and in the manufacturing
processes, like the links in a chain. The weakest link determines
how the product will eventually fail, so careful attention must be
paid to all. Recent trends have been toward a reduction of both
product failure rates and costs and further improvements can be
expected.
Area
Techniques
Design
environmental margin testing (Shmoo plots)
testing component stress analysis
life testing failure rate estimation
setting goals anticipating abuses
Materials
stress testing
component
screening
failure analysis
Manufac- sales inspection
turing warehouse audits
warranty analysis
control charts
process change reviews
incoming inspection
vendor performance tracking
specifications
production failure analysis
environmental requalification
process control
metrology
Service control of
control of mean time
turnaround time to repair (MTTR)
delivery time of audits of instruction
spare parts
manuals
customer training information feedback
Paul Baird, HP's Corporate Assurance Engineering Man
ager, is involved in product assurance R and D and in
coordinating product assurance activities at HP's far-flung
divisions. He is a senior member of the American Society for
Quality Control. With HP since 1961, Paul was involved in
voltmeter design for many years. He has a BS in mathema
tics from Oklahoma State University (1950), an MSEE from
the University of Colorado (1953) and an Engineer degree
from Stanford University (1959).
Bulk Rate
U.S. Postage
Paid
Hewlett-Packard
Company
Hewlett-Packard Company, 1501 Page
Road, Palo Alto, California 94304
HEWLETT-PACKARDJOURNAL
JANUARY 1979 Volume 30 • Number 1
Technical Information from the Laboratories of
Hewlett-Packard Company
Hewlett-Packard Company, 1501 Page Mil! Road
Palo Alto, California 94304 USA
Hewlett-Packard Central Mailing Department
Van Heuven Goedhartlaan 121
Amstelveen-1 134 The Netherlands
Yokogawa-Hewlett-Packard Ltd., Suginami-Ku
Tokyo 168 Japan
Editorial Director • Howard L. Roberts
Managing Editor • Richard P. Dolan
Art Director, Photographer • Arvid A. Danielson
Illustrator • Susan E. Wright
Administrative Services, Typography . Anne S. LoPresti
European Production Manager . Dick Leeksma
To change peels address or delete your name from our mailing list please send us your old address label (it peels off).
Send changes to Hewlett-Packard Journal. 1501 Page Mill Road. Palo Alto. California 94304 US. A. Allow 60 days
© Copr. 1949-1998 Hewlett-Packard Co.
Was this manual useful for you? yes no
Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Download PDF

advertisement