Discover more from The Asianometry Newsletter
Intel & AMD: The First 30 Years
If you want to watch the video, it is below
I want to take some time to pitch my Startup Island Taiwan podcast episode with Dylan Patel from SemiAnalysis. Dylan has been spending some time in Taiwan and I am happy that he took the time to sit down and chat semiconductors. Go check it out
At an Intel technical show in 2006, then-Senior Vice President - and now CEO - Dr. Pat Gelsinger logged onto a computer onstage for a demo.
His password was "I HATE AMD".
The rivalry between these two semiconductor giants goes back over half a century.
For this video, we will explore the first thirty years of that rivalry as the two moved from kinda friends to definitely not friends.
In the 1960s, most computers ran their logic functions within discrete, special-purpose logic chips. Everything was custom.
But it soon became clear that if we wanted practical, large-scale computer systems, then some standardization and specialization was needed. We cannot just be creating everything from scratch all the time.
Intel's original purpose at its founding in 1968 was to exploit the semiconductor memory market. They created special memory chips like RAMs and ROMs, chips that met a specific memory need inside the larger computer system.
This exercise had separated out the system's memory functions from its logic functions. So it was not a big step for Intel to consolidate those control logic functions into a single chip as well.
Thus in November 1971, Intel introduced the first commercial microprocessor - the Intel 4004. This 4-bit chip replaced six previous chips.
Originally made for a calculator, the 4004 set the founding principles of the logic microprocessor: Fetch instructions, execute them, and then store the results.
An advertisement at the time said: "Announcing a New Era in Integrated Electronics".
Over the next few years, Intel delivered new iterations of this device - raising throughput and cutting cost.
These include the 8008 in 1972, the 8080 in 1974, and the 8085 in 1976. A notable achievement was the 8080. Intel's second-generation microprocessor and the first designed from the ground up as a microprocessor.
Designers quickly adopted it into their products. The Altair 8800 microcomputer kit for instance was one of the first commercially successful personal computers. Bill Gates wrote Microsoft's first product - a BASIC interpreter - for the Altair 8800.
Other notable second-generation microprocessors of the day included Zilog's Z80, Motorola's 6800 ... and AMD's Am9080.
In 1975, a year after the 8080's release, AMD started producing its unauthorized second-source clone of Intel's market leading 8080 microprocessor: the Am9080.
This was part of AMD's early strategy. At the start, big buyers like the military - and later GE, Honeywell, and IBM - were reluctant to use a particular chip unless they knew they could buy it off multiple sources.
There were real reasons for this. One of the reasons Fairchild ended up the way it did was due to shoddy quality of work - micro-cracks in the chips. Second-sourcing also gave the buyers more vendors to buy from, adding a bit of price competition.
So AMD in its early days sought to be a reliable, high-quality second-source provider for those buyers. For more on AMD's start, go check out my video on the matter.
Often times, AMD did this second-sourcing without authorization from the original chipmaker. So they obtained the designs for that chip through reverse-engineering.
AMD's reverse engineering of the 8080 started with twins Kim and Shawn - later Ashawna - Hailey. They were reverse engineering enthusiasts, having picked up the hobby from a colleague at Xerox.
The twins took some 300-400 photos of a de-lidded pre-production Intel 8080 and shopped it around Silicon Valley at the time. They came across AMD, who was very interested. AMD gave them a team and a year later, they produced the Am9080.
The Am9080s were cheap to produce, about 50 cents each, and sold into the military market for as high as $700. It was a great business, and AMD became the largest seller of 8080s next to Intel itself.
Intel was not particularly happy about this unauthorized reverse-engineering.
But back then, the law held that computer hardware - in contrast to software - can only be patented rather than copyright-protected. And with patents, it is easy for things to devolve into outright, unproductive patent warfare.
Then in 1976, Intel announced that they would be installing microcode into their chips. Microcode refers to instructions at a machine code level. They are fabricated as part of the chip and directly control its functions independent of the operating system or the application software.
Today, we would probably call it firmware.
At this time, people were not sure whether microcode should be considered software or hardware. It had features of both. But Intel - with AMD's support - considered it to be software, and rushed to copyright it.
The presence of copyrighted microcode made unauthorized second-sourcing much harder. The Chip Protection Act later banned it entirely for commercial use in 1984. Thus, AMD and Intel struck a patent cross-licensing arrangement in 1976.
The two companies then had a somewhat brotherly relationship. Intel founder and Chairman Bob Noyce and AMD CEO Jerry Sanders were good friends. The two supposedly shook on the 1976 deal during a party at Sanders' house. Sanders long looked up to Noyce and acknowledged Intel's leadership.
In addition to the 8080, AMD would go on to second-source the 8085 with Intel's help. After that, they moved on and struck a second-sourcing agreement with Zilog a few years later for their Z8000 microprocessor.
AMD had high hopes for this product, but it failed to meet sales expectations. Largely because of the presence of another chip - the Intel 8086.
Signing the 1982 Deal
Released in 1978, the 8086 was the first Intel chip to come out with microcode.
It started off as a temporary fill-in. Intel originally was seeking to build what it called the 8800-series, an ambitious "micromainframe" type of processor.
When that 8800-series fell behind schedule, they rushed out the 8086. It sold well, but like I said, big customers would not accept the chip without a second source.
With semiconductor production capacity getting increasingly expensive at this time, this wasn't so easy. At the start, Intel had an early domestic source for the 8086 - Mostek out in Texas.
But then in 1980, Mostek's CEO informed Intel that they were dropping their second-sourcing agreements and going with Motorola - then Intel's arch-nemesis.
A domestic second source was necessary. At around this time, Intel had just won a contract with a renegade group within IBM for a "toy" product that would eventually be the Personal Computer or PC.
Released in August 1981, initial demand for the IBM PC far exceeded everyone's expectations. People were put on waiting lists for months. The IBM purchasing department then got involved and pushed Intel for a domestic second source.
Intel had signed agreements with Siemens in Germany and Fujitsu in Japan, but those companies could not build the 8086 or the similar 8088.
The whole ordeal was so annoying that Intel decided to partner with a second-source on a more ongoing basis. IBM suggested AMD, but as I mentioned AMD was already working with Zilog, a competitor. So Bob Noyce went over to meet Jerry Sanders to put out feelers.
In October 1981, AMD and Intel announced a second-sourcing deal that essentially extended the earlier 1976 patent exchange agreement. This new deal would last for another ten years starting in 1982.
As conceived, it was a mutually beneficial exchange of products based on those products' complexity and design effort - as defined by a formula. The goal being to make sure that Intel had instant second-sourcing without giving away the whole cupboard.
The second-sourcing agreement was supposed to last for a decade. But just a few years in, troubles started brewing.
The agreement covered the 8086, which was easy to handle. AMD wrote a check and Intel transferred over the product for second-sourcing. No problem with that.
In 1984, Intel wanted AMD to second-source their 80186 and 80286 chips, which weren't super-successful products but Intel had to get it out there.
But when doing the exchange, Intel found that there wasn't anything that they could get from or wanted on AMD's side. So they modified the deal to "loan" over some "complexity points" that would be paid back via higher royalties.
This was the beginning to a growing feeling that AMD was getting more out of the arrangement than Intel was. AMD got licenses for increasingly valuable products like the x86 microprocessors while in return Intel received licenses for peripherals with little market potential.
In October 1985 came about the Intel 80386 or just the 386. I will use that name throughout this writeup.
The 186 and 286 weren't the best products on the market, but the 386 was a whole different story. It was not only faster than the 286, it also had a 32-bit address which made it capable of handling the higher memory demands of a PC graphical user interface.
The 386 powered the Compaq Deskpro 386, a powerful PC clone that launched the entire industry and wrestled away the personal computer space from IBM.
Another major move that Intel took at the time was its decision not to offer second-sourcing for the 386. Internally controversial, it was a major strategic decision in semiconductor history.
Semiconductor economics were changing. Wafer fabs were starting to cost a whole lot more, and it made increasingly less sense to just be giving designs away to competitors.
There were two major reasons pushing Intel towards that decision to single-source the 386.
The first was that IBM wanted to take their microprocessor fabrication in-house. They were not interested in the 386, having created their own design based on a license they received for the older 286.
The second reason was that AMD did not deliver anything that Intel wanted in exchange for the 386 design. Company representatives went back and forth.
AMD argued that their products were perfectly good and that Intel had "Not-Invented-Here Syndrome", resisting products just because they weren't designed in-house.
Intel, on the other hand, believed that those AMD products were worthless and that AMD was producing them to be overly complicated just to game the complexity factor equation in the agreement.
Did I say two reasons? Should be three. The 386 was going to make a ton of money.
Fueled by the growing PC clone industry, being the sole source of the 386 would generate over $1 billion in revenue for Intel. With gross margins in the 80-90% range. There was a lot of money at stake here.
A Deteriorating Relationship
The brotherly early relationship between the two companies melted away.
Sanders, ever the flamboyant salesperson, had bragged a bit too much about AMD's Am286 - the 286-compatible second source - and it irked the guys at Intel.
Andy Grove had taken over leadership of Intel from Bob Noyce in 1987, bringing a new, more hard-nosed attitude to the chip giant.
Things finally boiled over. Faced with an impasse, AMD invoked a clause in the contract to bring the dispute before an independent arbitrator over the 386 and the 8087, a math co-processor. Intel tried to avoid such a thing.
So finally in April 1987, AMD petitioned the California Superior Court to impel that arbitration, which the Court granted. Intel responded by terminating the 1982 second-sourcing agreement, with one year's notice.
The arbitration hearings lasted for nearly five brutal years.
Finally in 1990 the neutral arbitrator - retired Judge J. Barton Phelps - found that Intel breached the 1982 contract. Seemingly validated, that year AMD moved to sell its Am386 - a reverse-engineered version of the Intel 386 chip.
Intel claimed to have found out about this chip thanks to a funny coincidence. AMD had sent documentation about the Am386 to its director of marketing for PC products who was staying at the Sunnyvale Hilton hotel.
Side note, this guy was named as Mike Webb in the press, but I have seen comments on Reddit saying that this was a pseudonym. But I have also seen a Mike Webb from AMD in electronics media at the time. So I am guessing it really is Mike Webb.
Anyway, it just so happened that an Intel employee with the same name - a field applications engineer - was also then staying at that Sunnyvale Hilton. The two even checked out at the exact same time. The hotel received the Am386 documentation and forwarded them to the wrong guy - the Intel employee.
That was Intel's story for receiving the documents. A funny, crazy coincidence right? AMD for their part did not quite buy the story and suspected impropriety or espionage.
Anyway, Intel - now aware that its rival was bringing a 386 competitor to the market - shortly went after AMD for having 386 in the product name.
But AMD as a second-sourcer has went down this path before and a judge backed their assertion that "386" is a generic term - allowing the Am386 to finally hit the market in March 1991 or about six years behind its Intel predecessor.
This ruling caused Intel to ditch the number monikers entirely. They rebranded the chip to the Intel386, or i386 - which can be trademarked.
Two generations later, they would start calling the "86" series of chips as they were back then as the "Pentium". That should have been that. But Intel wanted to explore every legal path to prevent AMD from bringing x86 products to the market.
I want to pause here to talk about microcode again.
Remember what I said about microcode? Despite having certain aspects of hardware and software, Intel insisted that it was entirely software and thus protected by copyright. In 1989, a judge would validate Intel’s claim in a major court case.
All the way back in 1976, the Japanese computer manufacturer NEC signed a patent exchange agreement with Intel - similar to AMD's.
NEC later not only produced a 8086 and 8088 second source design, they also produced their own Intel-compatible microprocessor - the NEC V20 and V30. Intel complained about this. The agreement allowed NEC to copy the hardware. But what about the microcode?
In 1984, NEC and Intel went to court over this issue. After five long years, the court finally ruled that microcode can be copyright-protected, though NEC did not violate that particular copyright. Intel lost that battle, but they won the war.
The year after the 1989 NEC court case, Intel brought a microcode copyright lawsuit against AMD for AMD’s 80C287, a math coprocessor that works along with the older 286 processor.
A year later in October 1991, they also filed another lawsuit of a similar line against AMD in an attempt to halt sales of the Am386.
Two years later in 1993, Intel filed yet another microcode lawsuit against AMD for their Am486 processor. This chip was compatible with the Intel 486, the higher performance successor to the 386.
AMD had developed the AM486 using what it called "clean room" or "Chinese Wall" procedures. This was a way of independently reverse engineering a product that was previously used by the IBM PC clone-makers to successfully re-implement the PC’s BIOS. Intel sued anyway.
Are you keeping track of all these? Throughout the early 1990s, AMD and Intel had some six court cases against one another.
Three of those cases are concerned with microcode - microcode for the math co-processor, the 386 processor, and the 486 processor.
There are two more lawsuits filed by AMD against Intel in 1991 and 1992. The first was a federal antitrust action accusing Intel of trying to secure a monopoly.
The second was a state suit accusing Intel of breaking California's Unfair Competition Act by demanding patent royalties from customers if they purchased an AMD 386 and 486 microprocessor.
This particular case had big repercussions down the line for the Taiwanese computer and motherboard manufacturing industry. AMD intervened and managed to prevent Intel from forcing exclusive agreements onto these manufacturers.
And then we still have the AMD-Intel arbitration over the 1982 second-sourcing agreement too. Let us pivot back to that one.
If you recall, back in 1990 the neutral arbitrator found that Intel breached the 1982 second-sourcing agreement. However, it was truly a plague-on-both-of-your-houses type ruling.
Yes, Intel intended to frustrate the contract, reject any AMD product for exchange, and keep that intent secret from AMD.
But AMD had not done itself any favors by failing to keep up with the semiconductor trends. Intel wasn't lying when they said that AMD's products at the time were not something that Intel actually wanted.
Furthermore, AMD also breached the good faith covenant by providing to Intel an older modem chip while already selling a newer one on the market.
In a subsequent hearing in 1992, the arbitrator awarded AMD $15 million as well as a permanent, royalty-free, nonexclusive right to the Intel 386 chip. He also extended the 1982 agreement two more years.
This final ruling was intended to allow AMD to sell the 386. But Intel appealed the arbitrator's decision in the courts - arguing that he had exceeded his boundaries.
The case would then go through the legal system for several more years. Twists and turns abounded. The California Court of Appeals overturned the arbitrator's decision, but AMD appealed that and so on.
Finally in the Market
Going to court allowed AMD to eventually bring the Am386 to the market.
Which was good since the Am386DX model sold quite well, despite entering the market half a decade behind its Intel counterpart.
The chip was functionally identical to the Intel 386. But since AMD produced their 386-compatible chip using a slightly more advanced process node - 0.8 micron rather than Intel's 1-micron - it ran about 20% better, ran cooler, and used some 30% less power than its Intel counterpart.
By the end of 1992, AMD had produced 9.5 million of these chips, generating a billion dollars in revenue.
It might have been more had Intel not lowered prices on their own chips in the first quarter of 1992 - something they might not have done if AMD wasn’t around.
The Am386 established the company's reputation as a legitimate alternative developer of x86 chips. AMD leapfrogged Motorola to become the industry's second largest semiconductor manufacturer.
So yes, these legal battles were necessary. But they were also bruising, invasive, and expensive for both sides.
AMD had spent over $100 million on legal costs. At one point, nearly $25 million a year.
Business for AMD in the late 1980s and early 1990s was not easy. Huge layoffs and a near-bankruptcy in 1986. A troubled merger with Monolithic Memories in 1987.
Not being able to release the Am386 for so long had caused the company's stock to plunge to a low of $3.62 in October 1990.
And the more time AMD spent tying itself up in these battles, the further it was falling behind.
For two months, top executives like Jerry Sanders had to sit in a courtroom listening to testimony all day before rushing to the office for meetings.
For Intel, the California Supreme Court finally affirmed the arbitrator's 1992 decision in 1994 after two long years of appeals. Intel didn't really have any more legal routes.
Furthermore, the 386 and 486 chips were quite old by now. The rearguard action had gone on long enough. It was time to call it and move on to the Pentium.
As Intel's director of litigation said, "We were almost litigating over an obsolete technology."
Thus in January 1995, the two companies negotiated a settlement for all of their court battles. Intel would pay AMD $18 million for breach of contract, and AMD would pay Intel $58 million for violating an Intel patent.
Jerry Sanders had every reason to celebrate this massive global settlement by throwing one of his trademark big parties.
But the flamboyant, controversially high-compensated CEO spent his first night after the settlement eating a pint of low-fat strawberry ice cream in his San Francisco apartment. Forever, a complicated soul.
The global settlement gave AMD a perpetual license for the microcode in Intel's 386 and 486 chips.
But nothing else thereafter. No Pentium. No nothing. AMD survived the Intel legal onslaught of the 1990s. But it was now officially on its own.