Published: September 5, 2019
Updated: September 5, 2019
Collection: Intel Memory Systems
Catalog ID: 2683
Published by: Dane Elliot
Intel started life as a Semiconductor Memory Company in 1968 with a vision of replacing the established magnetic core memory with a semiconductor memory, not as a microprocessor company.
The first six products were memory devices. And although the first product, the 3101, a fast small 16 by 4-bit bipolar device intended for computer scratchpad memory, the real focus was MOS specifically P-Channel Silicon Gate for building large memory systems for Mainframe Computers. The next five products were MOS memories, the 1101, 256 by 1-bit static RAM and the 1103, 1024 by 1-bit DRAM. Less familiar devices were the 1401, 1402 and 1403 1024-bit by shift registers.
Getting the world of mainframe Computers to accept and use these products turned out to be a difficult task. The technology was not easy to use, and potential customers were not easily convinced of its efficacy. This presented a significant obstacle to Intel’s vision and financial success.
Memory systems or (MSO) began with a proposal from William Jordan and Bill Regitz made to Bob Noyce in 1971. Although there is some reason to believe Bob was already thinking in these terms, the proposal put the creation of Intel Memory Systems in motion.
It’s now clear just how important MSO was to the success of Intel becoming the semiconductor memory supplier that pushed core memories and all other alternative technologies to the curb. MSO delivered the proof the semiconductor memory technology was viable by designing multiple mainframe computer memory systems and shipping large quantities of Intel memory devices in the form of mainframe memory systems. Using the 1103 1k-bit DRAM as many as 10k devices were contained in a single memory system product.
This effort provided earlier revenue for Intel and convinced mainframe manufacturers to design their own memory system solutions. Without MSO, Intel might have failed at worst or at least delayed their rise to what they are today.
For 11 years, MSO continued to play an important role for Intel well beyond memory systems, designing microprocessor systems for clients and once again proving the worth of the new technology (microprocessors) to the industry.
This approach of showing products using Intel semiconductors helping customers learn how to develop their own solutions by providing training and applications engineers to assist set the business strategy for the company for the next 40 years.
Intel MS (Memory Systems) proposal to Robert Noyce
Written by William Jordan and Bill Regitz, but no copy to be found!
MSO: A STARTUP STORY
Thoughts from William F. Jordan
Intel had been founded in 1968 as a supplier of memory chips. It was intending to be a supplier for the main memories of computers, an application that had yet to be proven. At that time I was an engineering manager at Honeywell (Framingham, Mass.), designing and manufacturing magnetic core memory systems for OEM sales and for use in our line of scientific computers. Anticipating the technological evolution, we embarked upon the product development of a semiconductor main memory, utilizing MOS dynamic technology. We had an in-house semiconductor pilot line and worked with several outside manufacturers, including Intel. Our project engineer was Bill Regitz assisted by Hank Bodio.
During one of our many trips to Silicon Valley, at a low point in our progress, Bill and I went for a SF cable car ride to relieve our tensions. I remember saying: “Can you imagine, one hundred years ago, some guy told his boss about an idea to build a transportation system with five mile long cables? If he could get that running, we ought to be able to get the 1102 to work!”
Intel was the most responsive supplier and eventually, by 1971, with Intel parts, we built a system, connected it to our 516 computer, and successfully ran diagnostics .
Bill and I were active crusaders for semiconductor memory, and our favorite forum was the Solid State Circuits conference. We had presented papers, were on stage for panel discussions, and in after-hour bars for spirited arguments. In 1971 Bill was scheduled to present a paper. I was told by my boss ( a plated wire memory enthusiast) that I couldn’t go because of budget cuts. I took a vacation, paid my own way and traveled with Bill to Philadelphia. Bill informed me that Intel had offered him a job as a chip designer, but he preferred to do system design and he didn’t know how to arrange that. I said I know how, lets send Bob Noyce a business plan. Many of the Intel people knew us since we had collaborated with them for over a year.
Knowing that Bob Noyce was a busy guy, I needed to get his attention so I sent him a telegram to his home. That’s always dramatic. He immediately called back, I then sent him a six page, 5 year plan, written in longhand. He said he liked the plan, especially the IBM add-on business. At that time IBM had 70% of the word market, and Intel had no way to sell memory chips to IBM because of technology differences. The remaining available chip market (the “seven dwarfs”) was a jumble. Also favorable was the fact that add-on memories have a high margin. However, Bob rejected two parts of our plan. Regitz had to work as a chip designer, the 1103 yields were so bad, we would have no systems if yield wasn’t fixed. Also, we couldn’t be based in New England; he had lived through the problems of trying to run a division (Fairchild Semi) 3000 miles from corporate headquarters. I offered some flailing arguments about earthquakes and Bob patiently quoted statistics about New England snow, hurricanes, tornadoes, and cold weather.
I moved to California and purchased a house with the San Andreas Fault cutting directly through my property.
First I hired Tor Lund as OEM engineering manager. He was working in Intel’s Applications Research Department under Ted Hoff. Tor had been building memory boards, could design all types of systems, and knew his way around Intel and the local supplier industry. Then we hired Bob Sumbs as Marketing Manager to help set our business approach. We soon hired Steve Campbell as IBM engineering manager, and also convinced Hank Bodio to move 3000 miles to help Tor.
The IBM add-on business was a departure for Intel, since we were selling to banks and insurance companies. In addition to getting the products designed, and into production, we needed a separate sales organization, maintenance organization, and we needed to provide lease financing. Intel’s financial VP, Larry Hootnick, assigned Mike Gullard to be our controller, and to help us with lease financing and the associated accounting challenges. Our OEM systems business was also growing, driven mostly by our long term relationship to Univac Corp. Our expanding success was followed by moving to our own new building space of 100,000 square feet in Sunnyvale. By this time we had expanded to hundreds of employees. At the computer conference on April 2 , 1973, Gordon Moore announced that we were promoted to division status within Intel. Years later after Intel stopped shipping memory systems, the lease payments continued to arrive like grandpa’s annuity checks, the gift that keeps on giving.
MSD created with William T. Jordan as GM – April 1971
Organization begins life on the west coast
MS divisional status Memory Systems Division (MSD) – 1973
As you remember it was a tough sell in the early days of Intel MSO. We were fighting core memories which were less expensive at the time and had NDRO capabilities. The biggest problem was the Intel sales force that had no idea what a memory system product was and were not willing to be educated. At the time Intel had a great many independent Rep organizations working for them and this is who I concentrated my efforts on. They were hungry to get commissions and were willing to work with us as bird dogs to at least smell out leads and call for help when they found something interesting. This involved a great deal of travel on my part and also on other early members of the Intel MSO engineering staff.
At that time most of the sales were for individual boards or small systems for specific applications. One of the big sellers was the 2102 board that was made into a large system for a gov’t agency and also was used by a couple of medical system companies. We plugged along with random sales, including a system for a so. California co. which was doing machine tool control, until our first big break which was a system for Univac.
A funny part of this was I had sold a system to the Univ. of Mexico which we built, but I would not release it until we received a letter of credit to cover payment. They kept saying the payment was coming and in the meantime Hank Bodio and Dick Belanger kept giving me a bad time about selling to a phantom university.They even delivered the system to my desk!! Well when Univac said they wanted a first prototype fast the only system we had was the UNIV of Mexico system, so you guys wirewrapped the Univac interface to it and we delivered it within a week. So I turned from a goat to a lamb in one week!! The rest is history ==more to come.
MSD moves to Sunnyvale – 1974
IBM 135/145 – 1974
I first interviewed with Intel December 1973. Ralph Ungerman was interviewing at MIT and I talked my way into seeing him even through I was an alum, not a student. In California I interviewed with Federico’s group in Santa Clara and with Steve, Tony, and Bill Jordan in Mountain View. I turned Steve’s job offer down. In April Steve contacted me again and this time I accepted. By this time Federico and Ralph had left to start Zilog.
Tony Prieto was project manager for the 145 add on memory. It shipped around the time I joined MSD at the end of May 1974. Tony had expected me to work for him on the 145 but Steve put me on the 135 add on project. Tom Defranco was the project manager for the 135. He was laid off in the RIF that came shortly after. I became project manager for the 135. Bob Blanding was the project manager for the 155 add on project. There was a lot of cross-project assistance. Bob would help on the 135 and I would work on the 155. This included design, debug off site (we rented time to attach and debug the memory systems rather than getting the systems in-house), installs at early customer’s sites, and second-level support calls. I recall working 30 days straight that fall to meet deadlines. We spent a lot of time on the road – New York, Chicago, Tacoma. We would redesign interface boards over the phone and ship them to the test site. We modified frame hardware to get the boxes to fit the mainframe.
Around this time we hired Lew Adams, Paul Bonderson and Charlie Guy right out of college or close to it. Lew and Paul have gone onto executive positions in Silicon Valley.
After the 135 project I proposed a 370/125 add-on. We hired to staff the project and rented a machine and had it installed in the lab and did two-shift debug. This was at the same time as the 370/165 add-on project that Bob Blanding managed. Around this time Steve hired Ernie Poblicion and Bob, Tony, and I reported to Ernie. Seven months later Ernie was gone, Bob took his job, and I started on RAP reporting to Tor Lund. Bill McCalmont replaced Bill Jordan during RAP.
After RAP was scuttled summer 1978 I transferred to work for Bob Garrow in MIcroprocessor Systems Division. The transfer was almost delayed because MSD IBM products were to move to Phoenix and transfers out were discouraged to encourage a critical mass of folks to agree to go to Phoenix.
Thoughts from Paul R. Bonderson
I started at Intel I think March 1, 1976. I worked with Charlie Guy and Jim Goodman in Dick Hodgeman’s department. My first projects were correcting and documenting wiring harnesses on the In 7158 Memory for the IBM 370-158. My first design project was the interface cards for the In 7158-MP for the IBM 370-158 MP. We had to design the hold off circuitry for the refresh of our dynamic memories in that the IBM memory was static. Our control card would sneak in a refresh during a write cycle or a read-modify-write cycle but if we couldn’t get a refresh done in time we had to hold off the processor by breaking into IBM’s control circuitry. The main signal we cut and held off was “WMS-Busy” (write main store busy). It was real tricky. Pretty sick that I remember this almost 40 years latter. We also had to extend their addressing circuitry for the Storage Protect Keys and the Translation Lookaside Buffers in three address bit increments. IBM was not cooperating with us add on guys, it was pretty difficult. We were adding cards into their backplane. I probably learned more about computers, memories, channels, and dynamic addressing in those years years than any other time in my career or schooling. It affected the rest of my career. When at Sun Andy Bechtolshiem was telling me how he had invented dynamic addressing for the Sun systems. I hate to tell you IBM was doing it on the 370 and probably the 360’s back in the early 70’s. I use to just chuckle. At the time Intel did not have a 158 to do our design work on so we would rent machines from whoever we could and use their machines from Friday at 8PM until Monday at 8 AM. We usually paid them with Memory that was selling for about $250,000/Mega byte. Charlie Guy and Jim Goodman where the brains behind the design, I was the grunt that implemented it and drew it up. All timing was done with hand drawn timing diagrams (lots of cigarettes) using the timing off the spec sheets. Kept great note books at the time. Very tedious work. We would have all of our design experiments worked out and in the log book before we got to the site. Every hour was planned out. We would work through the weekend with cat naps and coffee.
We had many troubleshooting calls all over the world. Local support guys would work the issue then call the specialist. If the specialist couldn’t fix it then the engineers would get called in. Local IBM guys hated us but usually by the time we got called in the IBM specialist was there and we all cooperated to get the customer up and running. Probably the neatest, and hardest problems were intermittent double bit errors. The worst one was at PEPCO (power co. Washington DC) . they would get an intermittent double bit error may be once every two to three weeks. Nobody could figure it out, including me. So we borrowed Ed Metzler from your group to go out with me to see what we could find. We schmood the machine, we vibrated cards, we zapped the machine with static electricity. We did everything could but could not make it fail. So before we left the site Ed suggested we take samples of the dust in the computer room to see what was there. Low and behold right next to the CPU and memory was a microfilm machine for microfilming the company records. There was silver oxide in the dust that was being sucked into the memory frame. That along with the silver plating on the driver chip dip legs silver migration was occurring between pins of the drivers. Poor design on the driver, -12 volts and +12 volts were next to each other, a microscopic bridge would form, a failure would occur and the bridge would melt itself thus fixing the problem. Ed took a control board over to Santa Clara, duplicated the environment and filmed it happening under the electron microscope thus proving the problem. The fix was to scrape between the leads of the drivers and conformal coat the drivers. We did it to a set of boards for PEPCO and installed them–no more failures. That would have made a great episode for todays CSI shows!
I have no idea how many service calls I went on but there were a bunch. Furthest away was Helsinki Finland. Longest and worst for me was Chrysler Mopar in Deerborn Mich. That was the machine I (young arrogant engineer) broke and lost all the beer to Blanding on. since the system did not come up on Monday morning I closed down MOPARS main warehouse and they had to send all the union workers home with pay! I’m surprised I didn’t get fired! Did the same thing to the city of Toronto once. MOPAR is were we figured out WMS busy for the MP system. It took Charlie, me, Jim Goodman, and an IBM engineer to figure out what was going on. I’ll never forget that one.
Almost all service calls I went on were intermittent failures and tough to find. Lots of times it ended up being a tri-lead with a broken ground wire. We could usually find them by margining the machine and zapping it with static electricity. We had an IBM zapper that had different antennas that you would use to locate the frame then the backplane, the card and then a hooked antenna you would place around the tri-lead. If the tri-lead had a broken ground it would almost always fail if you zapped it. The bad thing about the zapper was that you might work for 20-30 hours troubleshooting, get a bit rummy from no sleep, then forget to turn off the zapper and pick it up to move it. Boy you would be amazed at how fast 25,000 volts pulsing on and off can wake you up! The other great tool for trouble shooting was a diagnostic engineer who I forgot his name. He would come out with us and write custom diagnostics so we could get the problem to duplicate. If you could get the problem to happen it was usually pretty easy to figure out. The tough job was getting the problem to occur. All fun stuff and a great leaning experience.
Worked on all sorts of machines including two trips to Champion Paper products on a 360-155, Intels first IBM memory. Don’t remember what we were fixing or doing but it was Lew and I. The machines I worked on were the 370-158, 158 MP, 168 (support only), 3031, 3032, and 3033 and multi processor (MP) versions of all of them. Then on to the solid state disk drive that wasn’t as fast as drum it replaced. We did the “Fast Path” and did all the controls with the AMD 2900 bit slice processor. Gosh I think we had a 144 bit long control instruction. That is when we hired Jeff Rothchild (one of the three smartest guys I ever worked with.) Jeff went on to be the VP of Engineering and CTO for Facebook and was one of the original guys to start Veritas. We took the access time of the solid state disk from something like 35 msec (worse than the drum it replaced) to some thing in the range of 30usec for a read. Cycle time of the 2900 was some thing like 30 usec with a 10 or 15 usec clock time which was lightning fast in those days. We intercepted the channel command and if it was a read, write, or read-modify write the bit slice took over and did the command in one cycle. If it was any other command it was passed off to the 8086 for processing. Than was a fun project, learned all about the channel and spawned some of the ideas I had for storage area networks. Imagine that!
The IBM group at MSD designed boxes for the 145, 155, 158 and the 168 The 168 memory card was the basis for the C2 project and we had a dedicated water cooled 168 to do final test. It was at the time the biggest PO Intel had had, over $30M. Gelbach did not like the project at all, but he eventually supported it. It was a very well implemented program driven by Greg Pasco.
This also laid the corner stone for the follow up programs with IBM.
Here is another thought for all these new young techies The first IBM add on we did was for a 370-155, The box measured 4’ x 4’ x 5’ high. Fully populated this box with the built in Power Supplies could hold 1 Megabyte(That is not a typo) and we sold it for $750,000. IBM at the time was selling their memory for $1.2 Mil.
Begin Offering IBM leases – 1975
I don’t remember the date we started leasing though it would coincide with hiring Steve Schlumberger and it was still generating monthly payments when I left Intel.
I most vividly remember selling the add-on memories to the leasing companies before we went direct. The two major companies in the US were OPM and Itel. OPM was a real trip. I think the initials stood for Other People’s Money. We once sent Eli Morowitz to New York to collect a payment they had been promising us. The payment was big enough that the interest on it for a couple of days more than paid for Eli’s trip. The founders were later found guilty of fraud and went to prison. They were submitting leases for non-existing equipment to the bank for financing.
I also had to educate the Executive Staff on Present Value of money and sell the concept of PV accounting. I remember lugging a small-suitcase sized portable computer terminal with an acoustic coupler to an Executive Staff meeting and running “what-if” scenarios on the Ross Systems time share PDP-10. Ken Ross had developed a spreadsheet tool that pre-dated Visi-Calc.
MSD Direct sales organization Established – 1975
First MSD $3M month – January 1976
Thoughts from Tor Lund
We also built a memory system for the Burroughs’s BSP. This was to be a parallel processor and was aimed at the scientific community. The project never took off. We ended up with two systems shipped, and then the project crashed at Burroughs.
Thoughts from Tor Lund
We also had a very successful DEC add on to the PDP70. It was designed by Don Olker and service really liked it at the beginning. They collected service revenue, but did not have to repair anything. Later when none of the customers renewed the service contract, not so well liked.
Univac 9400 –
Thoughts from Tor Lund
The first one was the new memory for the 9400 because Univac had lost the formula for how to make plated wire memory which was the original memory for the 9400. We used standard IN-10 boards and built custom interface to work with the 9400. The original box was shipped to Roseville, MN and to say that the engineers there liked the development was not true. One engineer pointed out to me that the cover on the power supply tray was not sturdy enough to be used as a step ladder. This required a trouble shooter on Bert Boshe’s staff to fly out to Roseville to calm the nerves.
Univac 9300 –
Thoughts from Tor Lund
The 9300 was a card system and adding the Semiconductor memory to the system extended the life for about two years.
Univac 1108 First Shipment –
Thoughts from Tor Lund
The big Univac project was the Univac 1108 memory system. It was used on their biggest computer and installed at factories and banks. This is the system which was plagued by loose particles.
1103 Soft error problem caused by Alpha Particles –
Thoughts from Tor Lund and Bill Jordan
The early parts had an infant mortality problem. They took some time to bring the reliability curve down to the bottom of the curve. We built a cache system for GE Timeshare which had a very strong cooling fan and the failures would start occurring in about three weeks. Solution, disconnect the fan and do system test at temperature.
Alpha particle problem; we had numerous project going on with Component Reliability and MSO. We even built a lead cave. It was a nice experiment, but it just proved that whatever source we were looking for did not do the deed. The problem was in the device itself. The glass passivation on the die emitted alpha particles which changed the state of the storage cell.
Great catch by the Device Reliability people, particularly Tim May.
1103 Loose particles problem –
Thoughts from Tor Lund and Bill Jordan
An Intel 1103 memory system was built for Univac, to be used on their 1108 computer system. The problem manifested itself by random bit failures in the field, not seen at device test nor at board test.
Univac discovered the problem. Whenever they moved a perfectly good computer, it became full of memory errors. Albert Yu had a hearing-aid like device that could listen for noises of particles rattling around in ceramic packages. He used it as a diagnostic tool; finding packages for autopsies, which lead to the source of the problem. He also measured the statistical incidence which determined that we could screen it out. His device wasn’t practical as a production screen. We employed the “Branford” vibrator (shaker table) to re-screen everything in inventory, in the field, and in process. The problem was caused in assembly. Numerous packaged devices with open tops were stuck onto a Styrofoam board during inspection. After inspection the boards were stacked up. The Styrofoam boards picked up residual bonding wire tails on their bottom side (electro statically) and later dropped them into the ceramic cavities below, to be sealed in.. Intel corrected that assembly process, but we were stuck with many units in inventory and in the field. Univac was furious since they paid extra for ceramic, thinking it was more reliable than plastic. Andy, Albert Yu, and Bill Jordan went to Univac for a mea culpa visit. It was the only time anyone ever saw a nervous Andy tremble. Later Andy hosted a celebration party with lucite paper weight party favors showing huge loose wires in the ceramic packages.
To compound/reduce (depends on what side you were on) the problem, Andy agreed to buy Univac computer system for the computer center. Larry Hootnick was writing memos to cover his butt, everybody knew you bought IBM systems
Thoughts from Tor Lund
We had a burning problem with the 2147 used in the IBM add on sold to IBM. Device Reliability with Mike Geilhufe solved the problem very quickly to the complete satisfaction of IBM and the C2 program continued. IBM C2 $30M contract executed
The C2 contract was a bailout for IBM. They could not deliver adequate amount of 168 memory to their customers, so in their attempt to not get into delivery problems, Intel MSD was chosen as a supplier for all their in-house 168s. The original memory in the internal systems were shipped to external customers.
The contract was for $30 M and had to be shipped in 6 months in the form of a BSM (Basic Storage Module). In order to ship a good product, we rented a water cooled 168 CPU which we installed in Bldg. 2 in Sunnyvale. After completion of the project, Computer Services used the machine to run business applications before the 168 was shipped to Phoenix.
Thoughts from Tor Lund
We also built a few custom microprocessor boards. The most successful being the German scale manufacturer. This was the first intelligent meat scale ever produced and we shipped thousands to Germany before they started their own production line.