[转帖] Highly Accelerated Stressing of Products With Very Low Failure Rates

版权属于作者

Highly Accelerated Stressing of Products With Very Low Failure RatesHarry McLean, Reliability EngineerHewlett-Packard Co. Vancouver DivisionVancouver, Washington Written 5/90 - 10/91Minor corrections and reformatting 6/01
BiographyHarry McLean has worked for Hewlett-Packard for the last twenty-three years with many different responsibilities. During the last four years he has been responsible for manufacturing product reliability improvements. Most recently (the last year and one-half) he has been responsible for selecting and implementing a better method of verifying that the division's outgoing failure rate has not unexpectedly increased. Previous papers written by Harry dealt with programming a Hewlett-Packard 1000F Mini-computer system to better manage a manufacturing facilities department. Harry received his BS degree from Northeastern University in Boston.
IntroductionThis paper is a focused implementation of ESS or Environmental Stress Screening in a highly focused use -auditing rather than screening. At the end of this article is a glossary of the unique terms that are used herein.The problem of improving the monitoring of our out- going product quality is an issue that we have addressed during the past four years as we have been lowering the overall failure rate. A newly released product has most of its major issues resolved by the time it is introduced and plans are in place to correct any other known deficiencies, however minor they may be. The problem is then to monitor the out-going product quality in terms of reliability. We have had a number of test plans in place but none of them was very successful. Our problem is aggravated because of the long expected personal printer life (in excess of three years) and very low failure rates. More than 20 printers operating for many months were required to provide the necessary confidence levels (>80%) in order for it to be released to customer usage. Since our printers use cut-sheet paper and have replaceable printhead cartridges, they require constant monitoring by someone during testing. The time to perform a life test has caused us to turn to innovative methods for determining the product's life and failure rates with at least the same confidence levels as before. A further complicating factor is that we have never had a PC board failure at room temperature but have had some during BEST qualification. This makes the in-house testing a bit suspect because it does not correlate with what our customers are experiencing. This paper will discuss the solution of monitoring our out-going product quality which meets our needs.Typically, products that have been screened or audited by stresses should not be shipped to a customer until the proof-of-screen confidence levels have been met. This proof-of-screen is done so that the manufacturer can be assured that the customer's unit will meet or exceed its published or design life expectancy even after it has been stressed. In our case, this requirement would mean that many months would pass before products could be shipped and the delay for the proof-of-screen test to conclude was unacceptable. Hence, the requirement was to determine a life test that would not be as time consuming as “conventional”, life tests. The decision was made for the HASA project to begin by stress auditing PCAs. PCAs represent about one-half of our field failure rate. The decision to begin by stressing PCAs was also reached because a typical printer has many parts with different coefficients of expansion. When a unit has failed, the failure must be isolated and root cause reached so that corrective action can be evaluated and implemented. This meant that we either increase our time to introduction of HASA into manufacturing or buy a second chamber-system. Neither solution was acceptable because we had not yet proven that this process would work with our products. Regardless of whether one chooses to stress PCAs or finished products, some form of proof-of-screen is required. There will be further discussion on proof-of-screen later in this study.
Proposed Test & TheoryIn HASA, each PCA is subjected to a stress environment which is tailored for that PCA type because a screen developed on one PCA may not necessarily be an effective screen on a different type of PCA. The stresses applied were; triaxial random vibration, temperature slew rates approaching 30°C/min, temperature dwelling, and power line voltage and frequency variations. Although the level of these stresses is much higher than that experienced in a customer's environment, it has been the experience of many industry leaders that use accelerated stressing that when failures are uncovered and corrected, field failures likewise are reduced/eliminated. The first PCAs which were tested via the HASA process were not designed with built-in testability nor were there any confidence that they were rugged enough to survive the harsh stresses. We were pleasantly surprised when the PCA's worked very well under the stressed conditions without any design changes. In future products HASA testability will be implemented during the design phase. Typically, if a PCA is stressed for some number of cycles in HASA and if it does not fail, it can be shipped to the customer provided that proof-of-screen has been done. Even if a PCA fails it can be repaired, analyzed for root cause and shipped with the same degree of confidence of a non-failing PCA. The proposal was to stress a number of PCAs in HASA until they all failed. This group of PCAs became the control lot. The failure time, in test stress cycles, would be tracked with each PCA. The "time to failure" versus "percentage of population" would then be plotted and a normal distribution hopefully would result. The information from this distribution would provide the minimum number of stress cycles to end of life as well as the standard deviation of the distribution of failure times and the maximum expected life. The expectation is that with a large enough lot, both normal as well as infant mortality distributions could be seen.Since the failures which are not infant mortalities are due to fatigue and fatigue is an accumulation of stresses, the test can be halted and resumed many times with the same stresses because of the proportionality which was derived from Miner's equation.where, D, the cumulative fatigue stress, n, the number of stress cycles, s, the stress, ß is empirically derived and its value is between 8 and 12.An example of how this proportionality would be applied is to crimp a component lead so that the cross-sectional area is reduced by 50%. The stress at this point would be twice the stress at the same point on a good part because n(2s)ß rather than simply ns ß . Simply stated, defective parts are readily uncovered without sacrificing much of the life of a good part because the equivalent stress fatigue accumulation on the weak part would be many times that of the good part. For more information on Miner's equation, see Lambert's papers listed in the attached bibliography.On-going Monitoring of the Production ProcessThe objectives used for the HASA project were defined before the project began.


Detection of major manufacturing or vendor process shifts that would cause the out-going failure rate to increase by twice the baseline rate
Detection of problems in-house that would cause shifts in the out-going failure rate ōf =5% before they become customer dissatisfaction issues
Tested products in HASA would be shipped
On-going reliability feedback for product improvement
Testing must be achieved without becoming an obstacle to the production process.
Addressing each of the above objectives, we first selected the appropriate statistical equation for sample size N,Since the sampling that we need to do is a small proportion of an ever-expanding population, we selected a sampling plan that would meet our primary objective. That objective is to rapidly detect any major process shift. We chose Equation 2 and its only relationship with the original lot is the standard deviation, s , and the shift from the mean that we wish to detect, d . Nowhere does the equation state that we must have a sample size that is some proportion of the overall population. This is ideal for this type of problem because we are dealing with a very large, growing population and we only wish to detect some predetermined shift in its mean. The sample size to be tested (audited by HASA stresses) is NOT a function of the total population but is a function of the failure rate present in the population and the shift which we wish to detect. The following equation was used to calculate the sample size. Please note that this equation is the same as Equation 2 because as t a approaches Z , t ß approaches Z ß . The variance s 2 for this kind of data (binomial) is equivalent to p q and D is equivalent to d . where, Z & Z are the values from a normal distribution with 100 a (100 * ß ) percentage of area above itwith a being the risk of rejecting the hypothesis that the mean had not changed when it really had not changed. ß is the risk of accepting the hypothesis that the mean had not changed when in reality it had changed. p is baseline percentage failure rate which is the same as the out-going failure rate,

q is 1-p,
D is the change in the failure rate which is to be detected with probability ß and,
N is the sample size over a period of time in days.


By using Equation 3, with different values of a , ß , p, q, and D, yields the various sample sizes N (see Table 1). The "Days" column indicates the number of days that it would take to stress the necessary sample quantity N by sampling at 64 units per day. The "#Defects" column are the defects caused by the increase in failure rate, D, that would reach the customer and cause the out-going failure rate to rise from p to P+D. The "# Defects" column was calculated from an arbitrary daily shipment level of 1,000 units/day and an assumption that the daily sample level of PCAs would be 64 units. This is calculated by the following equation.Eq 4. #Defects = Days D Ship LevelIf the sample size is lowered for a given failure rate p and shift D, the confidence is, therefore, lowered. For instance, if we use the first row in the chart as an example and halve the sample size (N = 1,111), maintain Z at 0.842, the resulting Z would be lowered to 0.62. In other words, the risk would increase from 20% to 27%, which is a risk in allowing a bad lot to ship (because the mean had shifted). This, of course, defeats the purpose of HASA as an early detection of major out-going failure rate shifts before products reach our customers. By the way, the last row indicates that if we had a major shift in our product quality (e.g., a failure rate shift from 1% to 3%) it would be detected with a sample size of 112 units in just under 2 days with a 90% probability.There is no "magic solution" as far as which choice in Table 1 is the best. Selection of the risk () and the probability () must include the sample size (N) because as a and ß get smaller, the sample size will increase. The amount of the large shift (p) on the out-going failure rate (D) is also important because anything below this number is being acknowledged as acceptable and will not be a source of customer dissatisfaction. In essence, one must carefully evaluate all of the parameters and their impact on profitability.
Screen DevelopmentAll of the stressing of the PCAs was done with the print mechanism, power supply, and head driver PCA located outside of the chamber. These assemblies, in turn, were connected via a set of seventeen-foot shielded extension cables to the PCAs in the chamber. The PCAs were fixture mounted in the chamber in a fixture which could hold two PCAs simultaneously. The fixtures were bolted to the vibration table top {the chamber floor) and there was space to stress sixteen PCAs at one time. The stress levels were developed by stressing two randomly selected PCAs until either ceased to perform at the Destruct levels (see Table 2). It should be noted that when selecting the hardware that will interconnect the hardware which is external to the chamber to that which will be stressed, select components (cables & connectors) which will meet or exceed the environmental extremes that will be used. The actual level selected for the HASA stressing was then set to a level that was far above the Operation and Design Levels but below the Destruction Levels (see Table 2). During the screen development, the PCA's under test were subjected to the environmental system maximums (Destruct column) and did not fail -a tribute to the design. Table 2 shows the stress levels for which the PCAs were subjected and subsequently the screen levels selected.


Figure 1 is a representation of the actual screen but please note that the AC line frequency and volts were cycled twice, once during each vibration cycle at each temperature extreme. Nominal, 120V, 60- Hertz power was applied during the 25°C soak and during the up-going ramp. At the end of the screen, a soak at 25°C was done at nominal power line with a "tickle" vibration of 3Grms in which the operation of the PCAs was checked.

Notes:



The keypad LED's alternately flashed at 140C and returned to normal at 130C.
The keypad LED's alternately flashed at 79Volts and returned to normal at 87Volts.
All of the qualification testing was done on single and dual axis vibration and as of the writing of this paper, I am not aware of any correlation between these and triaxial random vibration.
The vibration was turned on for two 2-minute dwells at each of the two temperature soaks. Each vibration on cycle was followed by an equal off time of two minutes.
Slew rate is measured with the thermocouple fastened to the PCA via a crimp lug, i.e., this is the product temp slew rate.






The Equipment The decision to use random vibration was made by studying published papers on the subject in which comparative studies of single, dual axis, and triaxial shakers were made. Additionally, users of this technology were interviewed and their conclusions carefully analyzed.Although there are many assemblies within the HP9000 computer and other system modules, here is a simplified block diagram and the equipment used in the HASA system (refer to Figure 2):



Envirotronics Model EV70LN2, LN2 based chamber.

Model ORS 300CE, Triaxial Random Vibration.

JC Systems Model 620 Environmental Controller with FastTRAC ® & HPIB.

Hewlett-Packard 9000 Series 300 Computer with Unix ® Version 7.0.

Elgar Variable Voltage & Frequency Source Model 751SL-12, 9012-1 Oscillator & T option.

Hewlett-Packard 3488A Voltmeter with (2) 44471A Relay Boards.



The LN2 based system was a requirement because of the high temperature slew rates. An equivalent compressor based system, if available, would have huge compressors, consume a great deal of AC power and be susceptible to relatively high compressor maintenance. The LN2 chamber does not have the shortcomings of the compressor- based system. An oxygen depletion alarm, located in the area of the system for detecting LN2 leaks, is also required. The chamber is actually capable of temperature slew rates in excess of 60°C/minute.The in-house designed modules for controlling the printers in serial, parallel, and RS422 modes are not included in the above list and they are not commercially available. The sixteen-channel power mux module which interconnects the Elgar to the printers under test was also designed in-house.All of the software for controlling the system was written in C on the HP9000 Series 320 computer over a period of one year. Each function was developed independently, i.e., there are separate modules for vibration, power, temperature, and printer control. Each can be run separately which helps enhance the step screen development work. All of the program modules access a master table with the values for soak times, ramp times, events, delays, etc. This, of course, simplifies the programming as well as the user interface which is in XWindows. An example of the vibration table parameters are listed in Table 3. All of the times in the table are in seconds.



The Results of the Initial TestingFor a period of one and one-half months, sixteen PCAs were subjected to a number of stress environments. Each of the different stresses were documented and they varied from the final stresses in terms of slower temperature slew rates, longer temperature soaks, higher vibration levels as well as longer vibration dwells. The far right-hand column values in Table 2 were the final selection of stress levels to begin auditing of production PCAs. During the one and one-half months, none of the PCAs failed the stressing, and this became a source of frustration because failures are needed in order to verify the stress screen design. All of the PCAs that were screened were predominantly SMT (Surface Mount Technology) in nature although there were about twelve axial feed-thru type components. The next step then was to attempt to "seed" defects onto some new PCAs to see if the screen would detect the problems.In checking with people that had experience with screen development, we found that they had very limited experience in seeding SMT boards. Unfortunately, vendors don't normally keep a supply of process defective parts that work at room temperature and fail at either high or low temperature or during the ramping which could be used to seed boards. One, therefore, has to be creative in obtaining boards or components that work at room temperature and fail during the HASA stressing. An attempt was made to seed boards by having our PC repair people create problems that would pass the PC testing but later could fail in HASA. The first batch of sixteen boards were stressed for 98 cycles with only two plug-in ROMs becoming unseated during vibration at six and thirty-two stress cycles respectively. The results were not very encouraging and a second attempt was made to seed samples. Once again the same people were asked to create problems such as remove the solder from critical leads on certain components, thump the crystal on one or both sides and reverse the polarity on two electrolytic capacitors which were in the 5V supply. All of these problems had to pass both the PC test process and a mainframe test. The results of the stressing on this second lot of sixteen boards were even more disappointing because no boards failed after thirteen stress cycles. Once again, the ability to successfully seed SMT boards eluded us. A brainstorming meeting was next held with the PC test technicians to see what we could do to verify that the HASA process would work in our production environment. We uncovered an area in our PC manufacturing process that did a 100% visual inspection of the boards while they were still in panel form but with many of the components already loaded. At this station, the operator can either accept or reject the board(s). If rejected, it is reworked or scrapped. Since the SMT process is not perfect and solder defects occur, an attempt was made to isolate a number of PCAs that were visually rejected for "no-solder", "poor solder" and the like. After these PCAs were located they were marked and sent to the test area after all of the components were loaded. Once again the boards passed both the PC test system and the mainframe test.In our production process we have a return path for boards that pass PC Test and fail during top level testing. These boards are returned to the PC Test area for retesting. If they pass they are returned to the Top Level Assembly area once again and labeled as No Trouble Found or simply, NTF. Twenty-two of these NTF boards were diverted and sent to HASA for evaluation. The results were dramatic -SEVEN boards failed within three stress cycles. Here are the details:



No solder on ROM enable lead on SMT part (One cycle).
No solder on one of the address lines (One cycle).
Intermittent solder short on I/O reset line (One cycle).
Intermittent solder short from 5V to ground (Two cycles).
Defective RAM (Two cycles).
Intermittent short from 5V to processor IC (Three cycles).
Undetermined at this point (Three cycles).

In the normal course of events these boards would have been returned to the final product assembly area and placed in a printer. Later, possibly passed the final product testing and subsequently shipped to a customer -a few potentially dissatisfied customers.Following the testing of the NTFs, a batch of 42 boards were returned from our customer field repair center. These boards had all been either shipped mounted inside of a printer or shipped directly to the repair center for usage in the repairing of customer units. At one time or another, during the previous four months, all of these boards were sent to the repair center board test system because they had all failed. All of the boards were found to be either NTF or intermittent by the repair center. Once again, the stress screen proved to be valuable in inducing the latent problem to patent. EIGHT boards failed within three stress cycles. Here are the results:



Defective ASIC (Application Specific IC).
Defective 110 communication buffer chip.
Defective processor.
Copper flakes under solder mask causing short circuits,
Defective character ROM.
Poor solder wetting on two assemblies.
Shorted Zener diode.


While all of the stress auditing of the above lots was going on, a group of fifty PCAs began a "life" test. The first group of sixteen PCAs was placed in the chamber.The objective of stressing these boards was to stress all of them until they all failed. From this data it was hoped that a normal distribution could be obtained with a small distribution "bump" early in life that would indicate infant or premature failures. During the first two stress cycles four PCA's failed. These boards were removed from the test, repaired, the parts analyzed and then returned to the "life" test. At the end of two and one-half months, the stressing of sixteen of these boards was stopped at 1,251 cycles because no predominate failure mode was detected. Throughout the "life" test, failures occurred and most of them were attributed to the same ASIC that was the predominate failing component from the field return PCAs. The failure mode for these parts was a phenomena known as “delamination”. The basic explanation for this failure mode is the separation of the epoxy case from the metal interconnects which in turn causes voids in which humidity can be trapped. Over time, the humidity will cause the part to fail. All of these parts were photographed using acoustic microscopy without opening the parts.A lot of fifty standard production PCAs was stress- screened next. Since the chamber can only accommodate sixteen PCAs, the number of stress cycles for each of the four lots was intentionally different. Two lots were stressed for six cycles, one for eight, the other for thirty-six. The only failure seen could not be sourced when driven to root cause. The remaining 33 PCAs were never stressed to 1251 cycles like the first sixteen PCAs.

Table 4 is a summary of all of the PCA's that were screened during the HASA investigative phase. From the chart it can clearly be seen that this type of process will detect the types of manufacturing process issues that it was designed to detect.







PCA Source

NumberScreened

# StressCycles

NumberFailed

# Cyclesto Fail

Results of Number Failed


Standard Production

16

520

0

7

Used to characterize system


Std Prod/Seeded

34

121

1

6

Bent lead under ROM


Field Returns & FieldNTFs

42

642

9

1 to 3

Delamination, shorts, poor wetting, and old ROM code


Production NTFs

26

96

7

1 to 3

No solder, intermittents


Std Production

17

1251

4

1&2

Included in 7 NTFs


Totals

135

2630

21



Table 4



ConclusionsAs observed by other authors regarding stress screening, the observance and subsequent discarding of failure data during stressing can be costly. Cases have been documented in which failures were explained away as either that "the failure environment was not within the environment that the customer would use the product”, or that, “we haven't seen that failure mode from our field data". Experience dictates that each and every failure needs to be driven to root cause so that corrective action can be decided upon. The corrective action may be to do nothing for whatever reason but the failure needs to be documented as well as the corrective action for future discussion and analysis. The documenting of all failures satisfies two needs, 1) The problems uncovered can be sent back to the R&D team so that the next generation products won't have the same defects and, 2) It provides a historical database from which to draw inferences and conclusions about the product.Here is a list of the conclusions that we have reached regarding the use of HASA:



HASAhas shown to be a very effective tool in detecting or precipitating latent defects to patent in a few stress cycles. All of the recorded failures in HASA have occurred within six stress cycles or within three hours of HASA testing. We will begin using six cycles and try to reduce the number of stress cycles over time from six to possibly four.

HASA should not significantly degrade or deteriorate the useful life of the product. Presently, there are seventeen boards that have accumulated 1,251 stress cycles. If end-of-life is assumed to be 1,251 stresses, then HASA takes 0.5% of the useful life. There is no indication that 1,251 cycles is close to the end-of-life. From this testing, a conclusion that can be reached is that our boards are very robust and under normal conditions should work without any problems for a very long time. 2a- from the above statement regarding life, the PCA's exhibit infant mortality and from this limited sample of NTF's the infant failure rate was 8.3%. It is highly recommend that ALL NTF's be subjected to HASA because of the high failure rate. These infant mortalities would most likely fail within the first few days or weeks of customer usage. 2b- the same seventeen PCAs that were stressed for 1,251 cycles were later subjected to additional environmental testing to insure that HASA had not introduced some latent failure mode(s). The results of this testing clearly indicate that any concerns were not valid since all of the units passed. Here is a list of the testing: 1) A mini-Iife RVT test for 1/10th life at room temperature 2) ESD testing from 2.5KV to 25KV, 3) Design margin which checks 165 possible cases at different relative humidities and temperatures, 4) High and low operating relative humidity, 5) Supersoak, which consists of a 65°C 90% relative humidity soak for 24 hours and, 6) BEST temperature and vibration.

The original goal was to test sixteen PCAs per work shift. This goal will be met because we can perform the testing, including loading and unloading the chamber, in four hours assuming six stress cycles. Over a two-shift workday, two batches of boards (a total of 32 per shift) would be taken from the PC area buffer and returned within the ten-hour shift.

another objective of HASA was that it would not be a bottleneck to manufacturing. Removal of the boards in the required quantities will not be a problem.

HASA needs to have the capability of detecting large manufacturing process shifts in a short time period so that the quantity of defective products reaching our customers is reduced. Statistically, the HASA process will detect a failure rate shift from 1% to 3% in TWO days. A larger shift of course would take even less time. A smaller shift, from 1% to 1.5%, would take 28 days because of the small number of boards that are being tested. Initially, the primary focus of HASA will be to detect large failure rate shifts. This entire statistical system dictates that if no change has been detected in two days that we are 90% confident that the outgoing failure rate has not jumped to 3% or higher. Conversely, we always have the power to detect large shifts but the power to detect 0.5% shifts increases with time until the 28-day mark is reached. At this point we have 90% power in detecting this small shift. See Table 1. With all of the development work behind us we now look forward to bringing on-line a second stress system to be used for product development work. The first new product stressed in the R&D system will be the first product designed at HP with this process in place. The product during the design phase will be stressed to levels far beyond those seen in HASA so that we will know of the product's robustness and will only be concerned about ourmanufacturing process shifts and vendor process shifts. With the R&D system in place along with all of the failure analysis and corrective action processes, the next generation of printers will have lower field failure rates than those currently in production. All of this is done in order to provide a highly reliable product to our customers which in turn will mean a lower field repair expense and higher profits.

The results of the pre-production stressing in HASA indicate that the process will uncover all of those issues that were anticipated. In the future we will be linking the HASA PCA database with the manufacturing and field database. With all of this in place the complete life of a PCA or product can be reviewed from its inception to its eventual scraping.Statistical Limitations The statistical calculations in this paper use the normal approximation to the binomial distribution. This approximation is valid when five or more failures are apparent in the data. Exact binomial probability calculations should be used when working with fewer failures.

Glossary of TermsBEST. The acronym for Board Electronic Strife Test. A combination of test conditions which are used to uncover design or marginal defects in a PCA before it is made in production quantities.

ESS. Environmental Stress Screening. A combination of stresses designed to stimulate a product so that latent defects become patent. Performed on 100% of the products.HASA. The acronym for High Accelerated Stress Audit. It is a combination of stresses designed to stimulate a product so that latent defects become patent. Performed on a sample basis.Latent. A defect which is hidden. In reference to this paper it is a defect which is "waiting to happen". Stresses can cause a latent defect to become a patent defect.LN2. Liquid Nitrogen. Used as a refrigerant.Omniaxial Random Vibration. Vibration which is quasi-random in nature with six degrees of freedom.Patent. A defect which is not hidden.PCA. Printed Circuit Assembly. Includes the basic laminated board and its components.Proof-of-Screen. After the development of the stress screen, a test must be performed to verify that a "significant" portion of the product's useful life is not consumed by the stresses. The significant needs to be determined by the manufacturer.ROM. Read Only Memory. Memory in which data can be read but not written.Slew Rate. As applied in this paper it is the rate of change of the temperature chamber measured with a thermocouple secured to the PCA.Solder mask. A coating applied to the board before soldering which inhibits the solder from adhering to the traces.



BibliographyMany thanks go to Valerie Wildman and Isabel Rosenblitt, our division statisticians for their patience and help. Craig Daniels who was the original project manager at the inception and began the work of sampling and confidence levels for our divisional product failure rate goals and to Dave Shelley, my current manager who has supported my efforts. I also wish to thank the many people who provided their talents in the editing of thispaper.R.G. Lambert "The Shock and Vibration Bulletins 46, 47, and 53". Naval Research Laboratory, Washington, DC. August 1976, September 1977, and May 1983, respectively.R.G. Lambert " ASME Fatigue Life Prediction Under Random Loading". 84-PVP-3, June 1984.Course notes from Dr. Gregg Hobbs. Westminster, Colorado 80030. (303) 465-5988.

Suggested ReferencesStephen A. Smithson "Effectiveness & Economics", 1990 IES Proceedings.Dr Gregg Hobbs "ESS Course Notes".Charles Schinner " BEST- Board Electronic Strife Test", Hewlett-Packard Co. San Diego Division March 89.Bailey & Gilbert "STRIFE Testing", Hewlett-Packard Co Ft. Collins, Colorado November 1982 Quality Magazine.Clifton Seusy "Achieving Phenomenal Reliability Growth", Hewlett-Packard Co Boise, Idaho Disc Memory Division.