A Little Semiconductor Related History

[mawyatt]

As requested, here's a brief tour of way back in our developments of the first silicon fully integrated RF/MW/MMW chip developments. This is open for all sorts of additional information for anyone wanting to contribute.

Chapter 1

Way back around mid 80s we were always requiring a command channel receiver which needed to be very tiny and powered from various sources including switchers. This set us out on a path to create a fully integrated single chip receiver, and realized that we couldn't use a traditional heterodyne architecture which led us to create a direct-down conversion or zero IF type. We also couldn't use a crystal, as they were much too big, and created a unique receiver architecture which utilized the transmitted signal as the LO by injection locking an on-chip oscillator to the transmitted signal. The injection locking process also allowed direct FSK and ASK demodulation, this architecture was much later patented (long since expired tho).

We started looking for a 3/5 semiconductor process that could be used to integrate this receiver, and realized we couldn't integrate all the needed functions and turned to silicon. Many "experts" told us that this was impossible, as IC based silicon wasn't nearly fast enough, didn't have good MW models, no simulators for an entire receiver and so on. Well we didn't listen and developed our own simulators based around core SPICE.

When Bell Labs heard about this project they offered up a research developing complementary bipolar process with 9.8GHz NPNs and 4.8GHz PNPs. The models for the bipolars were what we called "Extended Gummel-Poon" and employed extra devices "outside" the core model to handle the MW and other GP limitations. We also needed integrated inductors in silicon, and utilized a thick surface gold metallization with a high resistivity silicon substrate. Later this silicon process became known as CBIC-V2 and was employed by Burr-Brown and many others for their high performance analog devices. Rumor has it that ADI got their high performance complementary bipolar process from AT&T Lucent Bell-Labs, but don't know for sure.

The integrated inductors in silicon were also interesting as we needed higher "Qs" than what others (Meyer of Grey & Meyer) were developing in silicon. These were helped by the mentioned which gold metallization and high resistivity substrates, and we developed means to create silicon inductors that gave reasonable "Qs", recall something like ~15 back then. These inductors were done with square spirals (layout tools didn't allow 45 degree angles), and we had experimentally verified that best "Qs" were achieved by not filling the entire square, but just about 40%. This is because the amount of incremental inductance vs resistance and substrate loss decreases as one "closes" the square spiral, thus keep the spiral open for higher "Q" needs.

The receiver was developed in this silicon process and had many other unique features including a high performance band-gap reference based LDO that had very high frequency PSRR because we couldn't use any type PS decoupling as these caps were too big (the BG reference was published in the AT&T Lucent Bell Labs Journal). A single off chip small capacitor was required to offset compensate the entire receiver chain, we also had a high value resistor off chip since it didn't effect overall size (located next to cap), could have integrated this but might have had issues with noise and substrate coupling, so decided for off-chip.

The on-chip oscillator had a 5 bit DAC to control/trim the frequency, and we also needed to fine tune/trim the oscillator and decided to use a laser and developed a laser trim feature as well.

The 1st chip actually worked really well, and we went on to production and developed a means to actively measure the chips utilizing a Flexible Membrane Polyamide probe from Cascade Microtek and also a version which allowed active laser trimming. The idea was to make the probe "look" exactly like the substrate the chip was mounted too, thus the laser tuning would be accurate.

Later a customer wanted a flip-chip version of the receiver to reduce the Z axis due the wire bonds, the entire chip and substrate had to be under 1mm!!! We wafer thinned the chips and developed very thin substate, and created bumps on the chip pads by carefully smashing a wire bond onto the pad which created a ball type "bump" which was flipped onto a substrate with conductive epoxy selectively placed on the substrate. This flip chip imposed a problem for active laser trimming the chips after flipping on the substate. We developed a substate with tiny holes to allow the laser to "blindly" trim the flipped chip. This all worked but was very tedious and expensive, so these weren't produced in large  amounts.   

Along with the chip development we worked with Dr Ken Lakin at TFR to develop very thin high performance BAW filters in Aluminum Nitride. Ken was the leading authority in BAW technology and developed the small thin preselect filters for the single chip receiver, we had plans to integrate the BAW filter with the silicon receiver but that never materialized.

End of first chapter.

Best,

[mawyatt]

Chapter 2.

As one might imagine we also wanted to integrate a lot of digital signal processing onto the single chip receiver as well as a companion transmitter. Silicon bipolar technology wasn't going to support he level of DSP and Logic, so we needed a CMOS or BiCMOS process that could perform at the speeds/frequencies we needed. IBM was early in the development of SiGe BiCMOS technology and wanted lots of $ to participate, more than we could afford at the time.

John Roberts founded SiGe Microsystems and heard about us and came down for a visit, they had licensed IBM UHVD SiGe technology created at Burlington, VT (another fun story). He wanted to license the Silicon receiver technology but management decided against such (another fun story). John mentioned that Maxim had utilized them (SiGe Micro) to help transition the old Tektronix GS2 bipolar process into a SiGe version called GST2.

We contacted Maxim and arranged a trip out to Oregon and were met by various Maxim Executives including Robert Shear (VP), we arranged an exchange with our wafer scale active laser trim technology for access to GST2. We were going to develop the Single Chip Transmitter for the Single Chip Receiver, with later minds of integrating both into a SiGe Single Chip Transceiver, and later with on-chip DSP.

Our Inital GST2 test chip didn't turn out well, the pinch resistors were way off and lots of other issues prevented us form getting much out of this first test chip. In the meantime IBM became available, so we abandoned GST2 since our ultimate goal was a fully integrated transceiver with DSP, and this would require CMOS and GST2 was just SiGe bipolar.

End of Chapter 2.

Best,

[mawyatt]

Chapter 3.

Robert Shear of Maxim wanted an introduction to Dr Ken Lakin at TRF technologies and insisted we fly out. He also insisted a direct flight to Portland and not take the shuttle flight to Bend, Oregon where TRF was. Robert wanted to pick me up and drive from Portland Courtyard (where I stayed) to Bend, Oregon (TRF Technologies). During the long 2 hour drive Robert asked lots of questions, both technical and otherwise, later I realized that drive was an interview!!!

Maxim was interested in acquiring TRF and that's why Robert wanted the introduction.

In the meantime we had arranged by lobbying US Congress to supply funds to allow an US agency to "buy down" access to an advanced SiGe BiCMOS process at IBM. This was successful and had early access to the advanced SiGe IBM Processes at Burlington, Vt. Much later this concept of using USG funds to "buy down" access to commercial advanced semiconductor processes for strictly USG use by qualified contractors became the seedling for the Trusted Foundry.

With access to SiGe BiCMOS we had the means to create an entire RF/MW/MMW Transceiver with DSP, Logic and Memory on a single chip. During this development we became concerned about digital "corruption" of the sensitive receiver by means of substrate coupling, and developed a substrate isolation technology later jointly patented with IBM. As a backup, since this transceiver was very important for certain customers, we developed a concept to sub-sample the input signal with a ultra-high speed precison sampler, then utilize DSP to recover the signal (we had experience with this subsampling MW signals in ~1980 with a Real Time SA, another fun story). The general idea was to sample the input just before the digital clock edges, since this gave to most time after all the clock edge chaos creates to die down. Our high level resources were completely consumed with the SiGe Transceiver, so we employed Q-Dot in Colorado Springs to develop the high speed precision sampler in IBM SiGe BiCMOS technology. The sampler was based upon a switched emitter follower concept, and later became a product of Hittite which had acquired Q-Dot from Semtech and now ADI.

Since the Transceiver was highly successful (after a hiccup with the diode protection on the many on-chip LDOs), we never employed the sampler, however it went on to a successful product by itself as mentioned. The single chip transceiver was produced in a few flavors for various intended usage, and led to the concept of a chip set for the JTRS military radios which created considerable interest in our little research company and eventually led to our acquisition in 2006 (another fun story).

Anyway, we've had the fortune to work with some great engineers/scientist over the years and we'll stop here unless folks are interested in the mentioned "other fun stories".

If you have addition or informative details related please chime in, this is all from old memory and surely forgot things, or messed them up!!

Getting old sucks, but it's better than the alternative

Best,

[jwet]

That is a fascinating tale.  I reported into Robert Shear (through at least one additional level of management) at one point.  I'll have to think through mine and respond.

[jwet]

Ok- Ok, hear is one from here, it taught me how the semi business works... - "How Dallas Became Part of Maxim"

Vin Prothro was the founder, CEO and president of Dallas Semi for nearly 20 years.  He started working for TI out of school and was part of a group of engineers that left TI to found MOSTEK- this was unusual for TI at the time- no one ever left.  MOSTEK was started in the late 60's and had seen explosive growth with semiconductor memory and early LSI around 72-74.  This early memory lead was something Intel and others were enjoying before the Japanese swooped in and killed its profitability a few years on.  This was also during the calculator and digital watch era as small LSI functions became reachable economically.  MOSTEK was right in the middle of all of these trends and were doing well.  Prothro rose to CEO of MOSTEK and the company was doing well having weathered the memory and calculator tsunami's as well as a dull late 70's economy.  United Technologies, a Connecticut conglomerate, mainly a Jet Engine company was on an M & A tear in the mid 70's.  Besides Jet Engines, they did industrial stuff like Otis Elevator, Carrier HVAC and were trying to diversify, a GE Model.  They  purchased MOSTEK on a whim.  They immediately wanted to move everyone to Houston- if you know Texas, Dallas and Houston may be in the same state but are worlds apart.  Prothro and a bunch of MOSTEK engineers founded Dallas around 1984 (BTW-about the same time as Maxim started), mainly to get a fresh start, work on some emerging ideas and stay in Dallas.  They started with memory but the Japanese soon made this a terrible business.  They had sharp applications people, solid design and good fab and manufacturing people.  They found some good niches, from the digital watch work at MOSTEK, they were uniquely positioned to do the CMOS clock and battery backed memory used in all PC's.  These battery plus IC hybrids lead into a lot of unusual assembly processes- embedding batteries, etc- "Touch Memory" was a surprise success that found niche wins- tracking keys in car dealers and parking meters in Turkey of all places.  The PC "ZeroPower" RTC was an awesome business and they had a heavy share.  They grew into other segments including Micros and Telecom T1 carrier circuits- with lots of market pull with local telecom companies in the area.  In early 2000, without any history of health or heart issues, 58 year old Prothro had a massive, fatal heart attack at a board meeting for a charitable org he sat on. (McDermott).  Dallas was devastated.  They had no succession plan, Prothro was a wiry and fit, young CEO.  The Dallas CFO, Alan Hale, filled in temporarily. 

A decision was made for a number of reasons that Dallas should be acquired rather than just replacing the CEO.  They had been successful but had sort of plateu'd.  Since founding in 84, their revenues had risen to $500M vs $2B for Maxim in the same time. The problem was to find a friendly, culturally and technically compatible acquirer.  TI certainly had the funds and some interest but would have been a disaster culturally.  Maxim was identified early on but our CEO was a bit skeptical.  Maxim was an analog integration company with very high margins, high engineering content and was selective about what markets it participated in.  It was strong internationally with half of its business being non-US.  Dallas was primarily a digital company, had only fair margins but generally had a less disciplined product launch machine.  Dallas also used manufacturers reps and did most of its business through distribution.  They didn't have a strong handle on their market position and had no international presence. Maxim on the other hand had a direct sales and applications force, did a significant amount of direct sales to large OEM's and had much better market knowledge.  The one thing that tipped the scales was Dallas fab capacity and unusual manufacturing capabilities.  Maxim was still trying to dig itself out of a deep capacity hole it had dug for itself in the mid 90's.  Dallas actually had an overcapacity, it built up for business that never came.  If Dallas could be self supporting without dragging down Maxim's margins (accretive), could provide some FAB capacity over time and if the people were flexible enough, it might just work.

It was no secret that Dallas was foundering but few knew how badly.  If Maxim was going to do something, it had to be fast and secretive.  Jack Gifford, before Maxim's founding, was CEO of Intersil after co-founding AMD with Jerry Sanders. He was running Intersil in 1980 when General Electric acquired Intersil.  Jack Welch and Jack Gifford weren't a good match- Jack G. quit shortly after and went back to his family roots of farming. He founded a farming company called J Leal Farms that he ran from 1980 to 1984 when Maxim was put together from other ex-Intersil alums.  I mention the farm because it comes up later in the story in a humorous context.

This is the end of Chapter 1- stay tuned.

[RoGeorge]

Wow, thank you!
Amazing stories.
Subscribed! 

[David Hess]

... Our high level resources were completely consumed with the SiGe Transceiver, so we employed Q-Dot in Colorado Springs to develop the high speed precision sampler in IBM SiGe BiCMOS technology. The sampler was based upon a switched emitter follower concept, and later became a product of Hittite which had acquired Q-Dot from Semtech and now ADI.

I always wondered what Hittite was doing for sampling.

[schmitt trigger]

The dynamism of the US semiconductor industry, between the 1960s and early 1980s was the World’s absolute highest. Only in the late 1980s did the Japanese companies, with their laser-focus on quality and cost cutting, did they find a worthy adversary.
I will be watching this thread for additional updates.

[jwet]

How Dallas Became Part of Maxim
Chapter Two- Top Secret Due Diligence

In order for one public company to acquire another, there are a lot of procedures that must be followed to make sure that the shareholders interests are kept in front of mind.  It also requires extreme secrecy strategically.  If the public gets wind of a sale, its inevitable that the buyer will pay a premium above the market price to entice the acquiree board, etc.  Its not unusual to pay 20% over market cap.  This was the early days of Sarbanes-Oxley and there was a lot of scrutiny on company shenanigans.  In order to preserve shareholder value for Maxim, a complete analysis of what Dallas really was was required.  Maxim needed to interview all the VP's, business unit heads, operational heads, financial execs and all the IC designers and senior apps and strategy people.  This was about 500 people in total.  This all had to be in complete secrecy to preserve the equity of both companies.

Gifford went to college on a baseball scholarship at UCLA and still played baseball for the "Maxim Yankees" in a serious adult league in San Jose.  Every now and then, I would run across some very buff 30 year old executive in the halls of Maxim- perhaps with a title of sales manager or expediter.  With a wink, a senior staffer in the know, would say, "Yankee".  Jack was recruiting ex athletes for mid level positions with the understanding that saturday afternoons would be baseball day.  These were all sharp, driven people but I found it funny.  Jack hired a lot of athletes for business management and junior executive positions.  Jack thought in sports terms with high energy and well thought out strategy.  This secret due diligence effort was a good example.  One afternoon, he took a chartered Lear Jet full of his senior executives down to Dallas to have a look at the place.  This was deliberately a little sloppy and overly public.  It was even leaked to the press outlets like EE Times and EDN.  The group did take meetings in Dallas with senior execs for a couple of days but the real purpose was strategic.  On return, the group settled back to work in Sunnyvale.  The press was anxious to hear the outcome and rumors swirled.  Jack and his team hadn't actually made up their minds, but the public word was that an acquisition didn't make much sense, there wasn't enough synergy, blah, blah, blah.  This calmed the gossip mill, but internal deliberations went on.  As the main product strategist for standard products, I was given several tasks to analyze product lines, financials, R&D spending and rates of returns of three of 24 of Dallas's business units.  There were about ten other senior staffers throughout Maxim that did similar secretive work after the well publicized exec trip.  We had about a week to complete a report that was forwarded to Jack's office.  All the rumours about any acquisition of Dallas died down.  Inquiries were answered with a probable no or a shrug in the rumour mill.
   Maxim had a lot of remote design centers and just lone remote employees.  This was somewhat unique at the time and shows a kind of flexibility of management.  I was recruited and hired in San Diego as the Field Applications Manager for the southwest.  I had been working in instrumentation development for 12 or so years at General Atomic and SAIC.  Good jobs but limiting.  I was always curious about working for a chip company and Maxim contacted me on an especially off day- the call came from Dave Fullagar- a founder of Maxim and Sr. VP who ran field applications.  If you know semi history, he's also the designer of the legendary 741 at Fairchild in the 60's, the part was defined by none other than Jack Gifford.  I was a bit starstruck.   I had interviewed with a couple chip companies upon graduation (ironically Intersil) but no offers came. I wanted to be involved in a chip company in the product selection, strategy and business development area.  In Maxim, this was called "corporate applications" and it sounded perfect.  Unfortunately, there were no openings and its not a job where a guy could come in from outside and be effective- you had to grow up some in the culture.  They offered me a job in Field Applications as a regional manager for the southwest, a pretty hot area.  From San Diego, I was to meet all Maxim's customers in the southwest, provide FAE support and hire 4 guys to cover San Diego, Los Angeles, Orange County and Phoenix and train them up.  Once I had this set up, we could talk about a corporate applications job.  Fair enough I thought and went to work. I had completed my hiring and training tasks after 18 months and was ready to talk about the next job.  My wife was from Tennessee and never was really wild about the west coast.  She put up with San Diego as its kind of heaven but she really wanted to get back to her roots in the southeast somewhere with her toddler family of two for all the normal reasons.  I looked around and settled on Raleigh-Durham, NC as a good candidate location.  I told Maxim i was NC bound somehow and would love to continue with Maxim if they could accommodate it.  We negotiated a remote corporate applications position for standard products where I would also cover the growing RTP Area as an FAE at least temporarily.  Similar to San Diego, I would hire and train a replacement and see where it went from there.  RTP was covered by a guy based in Florida that had all the southeast, a sizeable chunk of business extending from Fla to DC and west out through Alabama and northwest up to southwestern VA.  He was happy for some relief, RTP was about 1/4 or his territory's business but was growing fast- Ericsson Mobile Phones, IBM PC Company/Lenovo and Northern Telecom among many others.
    What was good in this situation about having senior staffers located remotely from the factory was stealth.  There were a couple dozen guys squirreled away that didn't really show up on org charts in a direct way. These people could be tasked to work and even travel for special assignments without anyone knowing.  On a friday morning, the CEO's admin, called me and said there was an open return ticket waiting at the counter at RDU for me for that afternoon.  Pack for a week- business casual or less- "you'll be a farmer"- flannels and jeans were ok.  This was pretty weird, Maxim was a very formal company- engineers wore dress shirts with ties, anyone higher was in a nice suit.  We were to check into the Four Seasons Las Collinas Golf Resort- pretty nice digs and be ready for a meeting at 9 pm.  There were about 75 people at the evening meeting, mostly remote staffers like me and some senior company muscle, our CEO and a bunch of finance types who I didn't have much contact with.  Las Collinas is about a 5 minute drive from Dallas in Farmer's Branch.  This Four Seasons would be our remote headquarters to take on the due diligence task.  Our cover story was that this was an executive retreat and golf outing for "J Leal Farms".  This cracked me up and I would throw random farming words into my public conversations to throw off any spies- "fertilizer!", "tomatoes!" became a private joke for years after.
     Over the weekend, we thoroughly interviewed 500 identified senior staff, designers and execs of the company.  Interviews were generally 30 minutes and key personnel would have three interviews.  Dallas employees streamed in and out of the hotel over that weekend.  Interviews started at 8am and went until 9 pm.  Some interviews lapsed into Monday with follow ups and clarifications.  All the interview evaluations were gathered, sorted and presented finally to the CEO on Monday night.  Each of us staff were given a single vote on whether we were for or against, I don't think each was weighted evenly but there was an illusion of democracy at least. The CEO and Sr. VP's deliberated privately that night and Tuesday morning drafted a memorandum of understanding that set a nominal purchase price and made our intentions public.  There was an a conference call with the investment community and trade press.  Later in the afternoon, a huge warehouse was cleared out that could seat about half of the employees.  Giant banners that said Maxim/Dallas magically appeared and all the Dallas employees were lead into a presentation in two shifts to hear about the deal and garner support.  All the Maxim staff was introduced briefly.  I was still in my "farmers clothes", not having the good sense to think ahead and bring some dress clothes.  There was a lot of excitement. Afterwards. all the exhausted stealth interviewers and staff disappeared back to their corners and it was over.  After such an announcement, there is a quiet period in M&A.  I found this unsettling after all the excitement.  Many months of travel to Dallas followed as we trained and met with our new Dallas colleagues.  Different Dallas groups were put under senior group managers at Maxim.  This went smoothly mostly, there were some inevitable, unavoidable overlaps that were handled pretty respectfully and the rest is history. 

The End

This was a great experience in my career.

[David Hess]

I really hate Dallas embedded battery backed up SRAMs and clocks.  They have caused so much trouble when the batteries died, or even ran low.  They are even worse than MOSTEK mask ROMs.

[jwet]

I can't argue much with your comments about the RTC/NVRAM stuff, they were loved by those that bought them but didn't age well.  They were not a very sustainable solution.  What these parts allowed was to plop a battery backed RTC and RAM into any standard 28 pin DIP JEDEC RAM/ROM socket to bring real time to anything.  Epson, I believe made similar JEDEC sized SRAM/RTC's with a replaceable coin cell on top.  They weren't near as popular.  Dallas carried on these non volatile RAMs up to a 128 Kbytes or more.  At the time of the acquisition, EMC made giant very elaborate and expensive RAID arrays where the allocation tables and RAID maps were backed up in a bank of these Dallas NVRAMs.  Since they were self protecting and battery backed, they were a nice backup solution for the day.  They must have been pretty reliable, EMC made pretty good stuff.  This was Dallas's single biggest design win and was worth millions.

All of this stuff has to be judged by the technology of the time I think.  ROMS and other larger parts of that ERA were notoriously unreliable.  This was later found to be sodium contamination in most early CMOS processes and Tin Whiskers that were the bain of early plastic packaging compounds.

It always killed me to open some fine expensive piece of test gear that had a Dallas RTC/RAM soldered down that held all the magic numbers and became EWaste after the battery went dead.  This disaster has to shared by the system designers.

[coppice]

Getting old sucks, but it's better than the alternative

Yeah, that Benjamin Button thing would be really creepy 

[SiliconWizard]

It always killed me to open some fine expensive piece of test gear that had a Dallas RTC/RAM soldered down that held all the magic numbers and became EWaste after the battery went dead.  This disaster has to shared by the system designers.

I second that. Pure waste, a great example of bad engineering, and just planned obsolescence ultimately.

I know some will (as always) argue that this isn't "planned" obsolesence as the intent was probably not to make it obsolete early, but when you perfectly know that a given implementation is going to fail after a given number of years, and you claim that it wasn't intentional, I'll call you either a liar, or a stupid cow. Pick your tag. It can be the cow if that makes you feel better.

[jwet]

A little serial eeprom in I2C or SPI versions existed at the same time.  A much better solution.  Self backed RAM was slightly easier for the system designer because he didn't have to design in some kind of last gasp power holdup to allow a final write for power failures.

I think the real lack was thinking ahead to the life cycle of that product and laziness.  Life cycle thinking should be part of engineering.  Laziness is just part of the human condition.

[coppice]

A little serial eeprom in I2C or SPI versions existed at the same time.  A much better solution.  Self backed RAM was slightly easier for the system designer because he didn't have to design in some kind of last gasp power holdup to allow a final write for power failures.

I think the real lack was thinking ahead to the life cycle of that product and laziness.  Life cycle thinking should be part of engineering.  Laziness is just part of the human condition.

They do think very hard about life cycle, but for a manufacturer life cycle typically means production date to end of warranty. 

For small amounts of stored data small EEPROMs were not only available in the late 80s, they were already cheap and massive volume. They went into every TV set to store things like the channel information. The life cycle issue there was that if you channel hopped a lot, through the adverts, those EEPROMs could wear out. I saw NVRAM used in a lot of places where an EEPROM would have been a better solution from the cost point of view.

[mawyatt]

As requested from # 2 (we had experience with this subsampling MW signals in ~1980 with a Real Time SA, another fun story) above, another fun story.

Around 1980 we were approached by a USG agency to develop a handheld battery powered MW RTSA that had been attempted and failed by two very well known and respected west coast technology companies. We were approached because of our advanced DSP in communications (High thru-put modems) capability and company in-house custom CCD technology which was needed for the custom CCD devices in a Chirp Z Transform based RTSA architecture.

How could a couple of redneck Florida guys do what the esteemed High Tek California Folks had failed at, not once but twice

The architecture was unique, employing the little known Chirp Z Transform rather than popular FFT approach to Real Time Spectral Analysis. For the CZT Sine and Cosine convolution we utilize a pair of very long custom CCD devices. These devices require a large voltage swing on 4 clocks to shift the charge down the convolvers, this alone would consume more power than was practical for a handheld device. One of the earlier attempts by one of big Califfornia companies utilized and patented a unique stepped transformer approach to reduce this CCD clock power, and not wanting to "Piss off the Wookie" we wanted to stay away from using such.

Realizing the charge stored in the CCD clocks was in the gate oxides and could be "redirected" and stored elsewhere to be used again, we are talking about many 1000s of picofarads for each clock and they needed to be driven to over 15 volts and often (high clock rate)!!! Do the math this was a big power drain just on the single Sine or Cosine CCD convolver and we ended up using 16 & 32 pairs in the final rendition!!

We came up with a technique we called Reactive Clock, where the clock charge was alternately stored in an on chip capacitor and reapplied. For this to work the CCD clock edge voltage followed a offset sine wave path from ground to clock peak, this was actually preferred for the CCD devices and improved their Charge Transfer Efficiency. A small switched inductor provide the means for this to work and was arranged in a fashion where the output voltage clock peak would double from 7.5 volts to 15 volts. So the clock would swing up to 15V from ground on the rise, be held at 15V, then switch down to zero on the fall and be clamped to ground, this would complete a clock cycle, with each transition when the inductor current was 0. The energy lost in the technique would be dominated by the NMOS switch and inductor loss, and could be managed. We ended up saving ~98% of the power lost in a conventional clock driver.

Fast forward to a lab test we were doing and with everything hooked up and running with 5 power supplies, 4 Tek Scopes and our custom AWG, a couple SA all running, we accidentally knocked off a clip lead to the 7.5V powering the Reactive Clocks. When reattaching the lead we noticed everything was still working without the 7.5V power, and the overall power was slightly lower without the 7.5V connected

How could this be possible?? Turns out what we had accidentally discovered was this unique Reactive Clock circuit would self servo to the lowest possible energy state without help!! Even compensating for CCD capacitance, inductance, switch variation, even temperature!! We should have patented this but didn't fearing the "Wookie might wake up"!! Much later we were allowed to publish so sent this circuit into Electronic Design (could be EDN can't remember) and Mr. Small was the editor, he said Bob Pease review it and said it would not work. We replied with a challenge to Bob, wagering a weeks of our salary vs. a day of his (he made way more than we did!!), and request he study the design a little more and maybe even breadboard it. Well it got published, and you can probably find it under Reactive Clock Driver

Edit: Here's a link.

https://www.eevblog.com/forum/projects/reactive-clock-driver-circuit/

Now back to the CCD devices themselves, they were in a special MOS process that was tailored for CCD imager use but also optimal for convolvers.  These were very long thin rectangular chips, way more than usually allowed for normal yields, and had a very large number of "taps". These "taps" were not the usual types but had special use for unique waveforms of interest which we can't discuss, but can say they are unique in they convey the impulse response of the convolver filter characteristics.

We'll stop here, and continue if folks are still interested.

Best,

[coppice]

The architecture was unique, employing the little known Chirp Z Transform rather than popular FFT approach to Real Time Spectral Analysis. For the CZT Sine and Cosine convolution we utilize a pair of very long custom CCD devices.

Were SAW convolvers ever considered for that project? They were all the rage in 1980.

[mawyatt]

The architecture was unique, employing the little known Chirp Z Transform rather than popular FFT approach to Real Time Spectral Analysis. For the CZT Sine and Cosine convolution we utilize a pair of very long custom CCD devices.

Were SAW convolvers ever considered for that project? They were all the rage in 1980.

Good question!!

These were very long convolvers, and SAW devices didn't have the equivalent Charge Transfer Efficiency we achieved with our CCDs, so with a SAW device the signal would likely have been severely attenuated . We were looking at just leaving a few electrons behind in each transfer with the CCDs, and the CTE was something like ~0.9999995 or better.

Best,

[T3sl4co1l]

How could this be possible?? Turns out what we had accidentally discovered was this unique Reactive Clock circuit would self servo to the lowest possible energy state without help!! Even compensating for CCD capacitance, inductance, switch variation, even temperature!!

You made a rail splitter with extra steps, so what?

Unless I'm misunderstanding the complexity of the situation, or it's just that no one else did at the time (that is, understand it, at a glance, on a sufficiently reduced enough level to see this particular kernel of truth)?

Either way, fun indeed.

It's kind of a shame that power transistors can't do this -- with gate drive, that is; I'd "invented" the same scheme myself years ago, but it doesn't work because internal R_G is too large to get any meaningful savings at reasonable switching speeds.  Unless you're going to drop Fsw massively (<10kHz?), which just isn't going to cut it for most any kind of power conversion application.

Not that transistor prices were ever low enough for that to be a practical route, anyway.  At least, you'd need a damn serious efficiency-per-buck requirement to have to do it.  And at that, with planar MOSFETs at the time (late 2000s) -- that is, requiring stonking massive dies for any kind of performance at high voltages, say for industrial gear.  Or to be a bit more concrete about it: say you needed to do synchronous rectification at 600VDC, at what point does your control power match/exceed conduction losses?  Then and only then, would you apply the resonant gate drive, and reap a little bit of savings.

Now that SJ are thoroughly dominant, performance is very nominal in all voltage ranges, and still continuing to improve from year to year besides; so the lossless gate drive idea is more irrelevant than ever, for power electronics anyway.

Ironically, SJ have their own unique loss mechanism, that limits output efficiency regardless of drive.  The thing with old planar is, on account of the ~quadratic voltage scaling limitation, you needed huge chips to do much power at efficiency, and lost relatively a lot as control power; well, SJ kind of does the same thing but on the output side (the SJ structure kind of looks like Coss dielectric loss, though it's a bit more interesting than that).  It's almost tempting to want planar parts again, to get high output efficiency in resonant applications; but the drive power, and general awkwardness of switching a transistor of some ~nF (big startup transients, or if timing briefly gets out of step), brings one back to reality.

Although that said, GaN often have quite low R_G, and I suppose given the small size, one might have a strict enough efficiency and size requirement, in a relatively unlimited budget (though, if that's aerospace: qualifications of those new-fangled GaN chips is going to be another matter), that it could pop up again.

RF transistors necessarily have quite low R_G, but they're never made in large enough junctions to get reasonable conduction loss in switching; or, not without paying truly exorbitant prices, say a handful of 1kW amplifiers just to switch a couple 100W.  Different fish kettles.

But doing that on your own chip, indeed the charge conservation and resistive losses can be optimized quite well.

Tim

[mawyatt]

How could this be possible?? Turns out what we had accidentally discovered was this unique Reactive Clock circuit would self servo to the lowest possible energy state without help!! Even compensating for CCD capacitance, inductance, switch variation, even temperature!!

You made a rail splitter with extra steps, so what?

Unless I'm misunderstanding the complexity of the situation, or it's just that no one else did at the time (that is, understand it, at a glance, on a sufficiently reduced enough level to see this particular kernel of truth)?

Either way, fun indeed.

Not exactly a rail splitter, but more of a rail creator since the Reactive Clock actually created the 7.5V supply and did so that it self servoed the lowest possible energy dissipation state compensating for variations timing, CCD capacitance, switches, inductance, temp, aging and so on. We could even use power from the 7.5V (and did) without affecting the overall Reactive Clock efficiency, since this stolen power was just another system parameter the Reactive Clock would compensate for, seeing it as just another inductor or switch loss.

Yeah it was so simple in operation at the time it totally eluded and fooled Bob Pease

It's kind of a shame that power transistors can't do this -- with gate drive, that is; I'd "invented" the same scheme myself years ago, but it doesn't work because internal R_G is too large to get any meaningful savings at reasonable switching speeds.  Unless you're going to drop Fsw massively (<10kHz?), which just isn't going to cut it for most any kind of power conversion application.

A decade later NASA patented a Power MOS Gate Driver based upon this Reactive Clock concept, we could have challenged the patent and shown "prior art" but decided not to pursue such since NASA was a major client of our company

 
Best,

[mawyatt]

RF transistors necessarily have quite low R_G, but they're never made in large enough junctions to get reasonable conduction loss in switching; or, not without paying truly exorbitant prices, say a handful of 1kW amplifiers just to switch a couple 100W.  Different fish kettles.

BTW in the 80s Honeywell developed a GaAs CMOS (another fun story) process, intended for Rad Hard applications. They also used the NMOS GaAs as switching devices for 1KW (5V at 200A) switchers for aerospace use and based upon GE's newly developed planar inductors, recall these ran at 200MHz. The general idea was to displace the usual 5V routing and place small 28V to 5V switchers at each load node, rather than a massive central power supply that distributed the 5V, analysis showed a significant overall system weight savings!

Mike

[David Hess]

Another problem with the Dallas backed up SRAMs is when the internal battery was exhausted, all access was lost even when powered, so it would not work even as a SRAM, which was fatal to most systems.  I had some motherboards that I really liked that used them, and when access was lost they would no longer boot at all.  I could not even repair them because the heavy copper in the 6 layer boards made even a through-hole DIP package too difficult to remove.

A little serial eeprom in I2C or SPI versions existed at the same time.  A much better solution.  Self backed RAM was slightly easier for the system designer because he didn't have to design in some kind of last gasp power holdup to allow a final write for power failures.

Supervisor chips to write protect and backup a standard SRAM chip existed from Dallas and others, and the circuit to do it did not take many HCMOS gates.

My solution was to use Autostor SRAMs from Cypress Semiconductor as drop in replacements, but they are mostly discontinued now.

[jwet]

Mike Wyatt-
I'll let you finish this tale and think of something else worth talking about, I hope.  You've got more technical stories than I do.  I was on the Apps and Strategy side working side by side with designers and business management to create new businesses.  Good stuff.

[RoGeorge]

if folks are still interested.

It's a delight to read such stories, please keep them coming. 

[mawyatt]

Ok, continuing from #17 above.

Another obstacle we encountered during development of the CZT RTSA was a means to sum a bunch of complex and fast Discrete Time Continuous Amplitude (DTCA) signals with high dynamic range. Conventional voltage feedback summing amplifiers were way too power hungry, and since the inverting summing junction is high impedance and only forced to low impedance by negative feedback which limited the ability to accurately sum signals as this required a "virtual ground" to sum properly at higher frequencies, we decide to start off with a low impedance inverting input, then let negative feedback drive it even lower!! This would come to be known a current mode feedback (maybe already was, but this was 1980), and we employed a current mode summing amplifier, first in bipolar then CMOS, we called "Summing Nodes", that consumed less than 10% of the equivalent voltage feedback methods to do the equivalent summing.

A second obstacle was how to test this CZT RTSA, the customer wanted to use NPR (Noise Power Ratio) as one test metric. The available instruments at the time employed fixed switchable high quality crystal filters for the  "notch" and amplified broadband noise sources for the "baseline", then Up and/or Down converted into the desired frequency band. These were extremely expensive and custom built at the time, beyond our incremental limited budget. So we came up with what could be called an Arbitrary Waveform Generator (AWG) which employed the fastest hybrid ECL 16 bit DACs available, a massive array of fast ECL Static RAM and a custom interface to the Apple II computer just introduced. The Apple was programmed in Basic to perform an Inverse Fourier Transform where the desired spectral waveform was drawn on the graphics screen, then transformed into the Time Domain coefficients and loaded into static RAM, which then ping-ponged the two DACs which had the outputs summed. The clock for the RAM & DACs were created with a high speed ECL programmable synthesizer clock generator. Also another higher frequency RF/MW/MMW generator was used to UP Convert the DAC outputs to whatever frequency band was of interest.

This concept worked beautifully except the conventional Spectrum Analyzers we had at the time had difficulty with some waveforms and we suspected (correctly) due to the discrete time waveform nature, so we set out to develop the sister to the AWG a Discrete Time Spectrum Analyzer based upon the highest speed 16 bit ADC hybrid available, another Apple II and more fast RAM and so on, then did conventional synchronized FFT analysis as required. This became critical, as did the AWG, pieces of custom test equipment for developement.

After all this we were able to create arbitrary spectral "waveforms" across large frequency spans in RF/MW/MMW ranges with excellent signal fidelity, enabling the mentioned NPR metrics to be gathered, as well as some very special "waveforms" we can't mention.

We delivered a handheld (size of a typical textbook) RTSA capable of spanning RF/MW/MMW frequencies that consumed under 4 watts battery power and had the equivalent computing power a third Cray 1 at the time

This opened the door and put us on the map in a very specialized community and we went on to many endeavors, many of which we can't discuss, but a few tidbits maybe we can discuss if interested, one of which was the first capacitance ratio ultra-low power SAR floating point ADC, but don't want to hog all the bandwidth!

Best