Fullscreen HTML5Med Res (??MB)HTML4

Part 1 by Dr David Maddison, VK3DSM Image source: https://freetrade.io/news/the-story-of-amd THE HISTORY OF AMD from garage startup to innovator We described the history of Intel in the February, March & April 2026 issues (siliconchip.au/Series/457) since Intel is one of the most important companies in the computer space. Its closest competitor, Advanced Micro Devices (AMD) also has a long and storied history that deserves to be told. I ntel was established on the 18th of July, 1968, while AMD (Advanced Micro Devices) was incorporated less than a year later, on the 1st of May, 1969. Like Intel, AMD’s fortunes have been made mainly on the back of computer CPUs. AMD’s market capitalisation (the total value of all outstanding shares in the company) surpassed Intel’s in February 2022. Today, AMD’s market cap is ~US$1 trillion while Intel’s is ~US$600 billion at the time of writing. The Fairchild Semiconductor connection Intel and AMD have some history in common, as they were both founded by former employees of Fairchild Semiconductor, and are thus known as “Fairchildren”. In 1955, William Shockley, winner of the 1956 Nobel Prize in Physics for 12 Silicon Chip co-inventing the transistor, founded Shockley Semiconductor Laboratory in Mountain View, California (primary operations began in 1956). Shockley was a brilliant scientist but a poor manager. In 1957, eight of his key employees, who came to be known as the “Traitorous Eight”, left to found Fairchild Semiconductor. Fairchild Semiconductor was not founded as a totally independent company like it might be today because the concept of venture capital funding didn’t really exist back then. Thirty companies were approached to assist in the founding of this new venture. It was Sherman Fairchild of Fairchild Camera and Instrument who provided US$1.38 million in funding (a very large sum at the time) plus a US$3 million buyout option. In exchange, the new semiconductor operation was Australia's electronics magazine established as a division of Fairchild Camera and Instrument, named Fairchild Semiconductor. This arrangement gave the eight engineers the capital and resources they needed, while giving Sherman Fairchild control and the potential for making a huge profit in the fast-­ growing semiconductor business. Fairchild Semiconductor became one of the most fertile companies in history, in the sense that many of its engineers and executives left to start other important firms. Such companies include: • National Semiconductor (1959) • Amelco (1961) • Applied Materials (1967) • Intel (1968) • AMD (1969) • LSI Logic (1981) • Altera (1983) • SanDisk (1988) The venture capital firms Kleiner Perkins (1972) and Sequoia Capital (1972) also emerged from Fairchild. Jen-Hsun “Jensen” Huang worked at both AMD and LSI Logic, both “Fairchildren”, and became co-founder of Nvidia, another huge technology company that we will discuss in a future article. Disillusionment Robert Noyce and Gordon Moore were weary of what Fairchild Semiconductor had become, so in 1968, they left to found Intel. A year later, Jerry Sanders, the worldwide sales manager at Fairchild Semiconductor, also became concerned about the dire situation there and left to found AMD. The disillusionment came from the mismatch between the fast-paced, rapidly developing semiconductor industry and the old, slow-moving, conservative Fairchild Camera and Instrument parent company on the opposite coast of the USA. The parent company also took much of the profit from the subsidiary, which starved the semiconductor division of funds to reinvest. Executives were also given poor stock options and reimbursement compared to other Fairchild divisions. Finally, the East Coast management rarely visited the West Coast company, and was out of touch and bureaucratic in contrast to the entrepreneurial, hands-on approach of the semiconductor division. siliconchip.com.au The dissatisfaction grew noticeably through 1966-67 and reached a breaking point from late 1967 to 1968, leading to Noyce and Moore leaving, with Sanders following less than a year later. Unlike Noyce and Moore, Jerry Sanders was not a member of the Traitorous Eight that founded Fairchild Semiconductor. He had joined Fairchild as a sales and marketing executive, where he quickly rose through the company. However, due to management conflicts, he was pushed out of day-to-day operations and eventually took a severance package of one year’s salary. He was known for his charismatic but controversial style as director of marketing, his flashy style, his pink Cadillac and his aggressive sales approach. Jerry Sanders founds AMD After being pushed out of Fairchild in late 1968, Jerry Sanders joined a group of seven other ex-Fairchild employees who wanted to start their own semiconductor company. On the 1st of May, 1969, the eight of them officially incorporated Advanced Micro Devices (AMD), with Sanders as president (he insisted on the role, which caused some initial tension). The company initially operated from temporary premises in Santa Clara, California, moving to Sunnyvale later that year. The original eight co-founders were (as shown in the lead photo & below): 1. W. Jerry Sanders III, from sales & marketing at Fairchild, became President & CEO of AMD 2. Jack F. Gifford became director of marketing and business development at AMD 3. Sven E. Simonsen became director of engineering, complex digital operations at AMD 4. D. John Carey became managing director of complex digital operations at AMD 5. James N. Giles became director of engineering, analog operations at AMD 6. Edwin J. Turney became director of sales and administration at AMD 7. R. Lawrence “Larry” Stenger became managing director, analog operations at AMD 8. Frank T. Botte became director of development, analog operations at AMD As you can see, they came from sales, marketing, engineering and operations backgrounds at Fairchild. AMD started with almost no capital. Sanders personally raised roughly US$1.5 million through a combination of private investors, small venture capital groups, personal loans and friends. Unlike the Traitorous Eight (who had corporate backing from Fairchild Camera and Instrument), AMD was a true “bootstrap” startup. The company’s original business plan was simple and pragmatic: they wanted to become a “second source” supplier of digital integrated circuits. They would legally copy (or license) popular logic and memory chips from competitors like Texas Instruments, Motorola and National Semiconductor, then sell them at competitive prices with better customer service. AMD’s first product (the Am9300 shift register) shipped in 1970, and the company made its first profit in 1971. The strategy to become a second-­ source supplier was very wise because at the time, big buyers of chips like the US Military, General Electric (GE), Honeywell and International Business Machines (IBM) were reluctant to purchase chips unless there were multiple sources for them. The reasons for this included that there might be quality, supply or cost problems if they relied on just one supplier. Imagine investing millions in developing a new product and the sole supplier of critical components files for bankruptcy! Identifying this market niche and becoming a high-­ quality, reliable second-source supplier for such buyers was an excellent move to help gain market share. “Borrowing from friends” AMD started with almost nothing. The company was incorporated with From left-toright, the eight co-founders of AMD: W. Jerry Sanders III, Jack F Gifford, Sven E. Simonsen, D. John Carey, James N. Giles, Edwin J. Turney, R. Lawrence Stenger and Frank T. Botte. siliconchip.com.au Australia's electronics magazine November 2026  13 roughly US$100,000 of combined personal savings and small contributions from the founders. Sanders then personally led an intensive fundraising effort to raise the US$1.5-1.75 million needed to actually launch operations. This was a very small amount for a semiconductor startup, even by 1969 standards. He borrowed heavily from friends, personal contacts and a handful of small investors. One important contribution came when Jonathan Lovelace (the founder of Capital Group) personally advanced US$30,000 just to cover minimum rent and basic equipment to keep the fledgling company going. Sanders used his salesman’s bravado to close the deal, repeatedly assuring the nervous co-founders that “we’re going to get the money” even when things looked uncertain. The funding phase was finally completed in mid-1969 after months of sleepless nights and near-misses. First headquarters – humble beginnings The first “headquarters” of AMD was in the living room of co-founder John Carey’s house. They then rented two small back rooms (about 84m2 total) inside a rug-cutting company at 1040 DeGiulio Avenue, Santa Clara. It was a shared industrial space with no proper amenities; just enough room for eight desks and basic drafting tables. The founders joked that it lacked even a garage door or convenient parking. Once most of the funding was completed by September 1969, AMD moved into its first permanent facility at 901 Thompson Place, Sunnyvale. This was a 1394m2 building they purchased for US$500,000. This space housed the entire company, including the first small wafer fab, which was set up for an eight-mask, two-inch (51mm) wafer process having a 7μm feature size. It originated from Sanders’ unsatisfactory experiences at Fairchild. He had watched the parent company (Fairchild Camera and Instrument) repeatedly mistreat its best engineers and salespeople by under-paying them, giving them poor stock options, and taking profits from the semiconEquipment & early operations ductor division to fund the failing East They improvised everything with Coast businesses. the absolute minimum of equipment, Sanders believed this “profits first” buying or leasing the cheapest basic corporate mindset destroyed morale semiconductor processing equipment and drove away talent. When he they could find, such as wafer process- founded AMD, he deliberately chose ing tools, assembly equipment, etc. the opposite approach. Starting as They also used a lot of second-hand early as 1969-1970, Sanders repeator reconditioned equipment wherever edly stated his new motto in numerpossible. ous early interviews, speeches and As mentioned above, they started internal meetings. with a rudimentary two-inch wafer What he meant by this was to hire line; nothing was state-of-the-art. the best people, treat them with respect They focused on second-sourcing, and generosity (such as better stock legally copying existing logic and options, bonuses and culture) and crememory chips from competitors like ate an environment where employees TI and National Semiconductor, so felt valued and motivated. He thought they didn’t need to perform expen- that if he did that, great products and sive research and development (R&D) strong profits would naturally result. right away. The saying became AMD’s unofficial Its first product, the Am9300, was motto for decades and was printed on released in 1970. It was a second-­ corporate materials, repeated in meetsource version of a product originally ings and used for recruiting. Sanders designed by Fairchild Semiconductor. credited it as one of the main reasons The whole operation was extremely AMD was able to attract top talent lean. The founders even did much away from much larger companies of the early work themselves, while during the 1970s and 1980s. Sanders was out raising money and AMD’s main historical eras “selling the vision”. AMD has had five main historiSanders’ motto cal eras. These will be used as main Jerry Sanders’ motto was “People headings in this article. The eras are first, products and profits will follow!” roughly as follows: It became his lifelong management • Second-sourcing, cloning, philosophy and one of AMD’s core reverse-engineering and some indecultural tenets from the beginning. pendent designs (1969-1995) Fig.1: AMD’s previous headquarters in Sunnyvale, California (since demolished in 2019). Source: https://w.wiki/RGex 14 Silicon Chip Fig.2: AMD’s Santa Clara, California, headquarters in 2020. Source: https://w.wiki/RGew Australia's electronics magazine siliconchip.com.au • Independent architecture and the GHz race (1996-2003) • Innovation leadership (2003-2007) • Dark years and the Bulldozer architecture (2008-2016) • The Zen renaissance and the modern era (2017-present) Second-source, cloning, reverse-­ engineering and some independent designs (1969-1995) Second sourcing was the foundation of AMD’s business model and the key reason the company survived its difficult startup phase. In the semiconductor industry of the late 1960s and 1970s, a second-source company legally produced and sold an exact or near-exact copy of another manufacturer’s integrated circuit. Customers wanted second sources because they reduced supply risk, created price competition and guaranteed chip availability. Being a “second sourcer” as a fundamental part of AMD’s strategy in the early years was very pragmatic. They deliberately chose not to compete head-on with giant companies by inventing new chips, which would have been very difficult for them at the time. Second-sourcing enabled them to relatively easily generate revenue to grow the company without a massive investment. Jerry Sanders’ famous by-line during this era was: “We’re not in the semiconductor business. We’re in the customer service business.” This second-sourcing phase was roughly 1970-1978, keeping AMD alive during its cash-poor early years. It helped build the manufacturing discipline and customer base that allowed the company to later develop its own innovative products. AMD chose to second-source popular logic and memory chips in particular. They reverse-engineered or licensed designs from companies like Texas Instruments, National Semiconductor, Motorola, Intel and Signetics. AMD produced functionally identical parts with the same pinout and electrical specifications. However, it tried to distinguish its products by offering better quality, better packaging, better customer service, or all three. As a result of its approach to second-sourcing products, AMD could offer customers: • High reliability (they built a strong reputation for high quality and low failure rates) • Better customer support, which came from Jerry Sanders’ background in sales, promoting an attitude of providing excellent service and response to customers • Military-grade parts; AMD focused heavily on the military and aerospace market, where customers were prepared to pay premium prices The second-source model worked because it required much less R&D than designing original chips. It allowed AMD to build manufacturing expertise and cash flow, and it created strong relationships with customers who wanted alternatives to the dominant suppliers. AMD’s earned reputation as the “high-quality second source” later helped them transition into original processor designs, starting with the Am2900 bit-slice family in the mid1970s, although that wasn’t its first clean-sheet product. A bit-slice processor is a modular, multi-chip building-block approach to designing high-performance custom microprocessors. Instead of creating a complete fixed-width CPU (such as eight bits or 16 bits), the Am2900 provided 4-bit wide ‘slices’ that engineers could combine, such as four slices for a 16-bit processor, eight for a 32-bit processor and so on. Each slice contained its own arithmetic logic unit (ALU, such as the Am2901 chip), registers and shifting logic, while accessory chips like the Am2909 handled sequencing and control. This architecture was highly flexible and microprogrammable, allowing designers to create custom instruction sets and optimised processors for specific applications such as minicomputers, graphics systems, military equipment and disk controllers. The Am2900 family became popular because it offered excellent performance and scalability at a time when standard off-the-shelf microprocessors were still relatively slow and limited. First second-source products The first second-source product, released in 1970, was the Am9300, a 4-bit shift register. After that, they quickly expanded into TTL logic families (7400 series equivalents), memory chips and linear (analog) chips – see Table 1. Another important second-source milestone was AMD’s first memory chip, the Am3101 64-bit bipolar RAM chip, released in 1971 (see Fig.3 overleaf). It was based on Intel’s first product, the 3101, released in 1969. It was not licensed from Intel but reverse-­ engineered. Table 1: AMD’s second-source products, 1969-1975 Year(s) Product Description Original source Significance 1970 Am9300 Parallel load 4-bit shift register, 5V TTL Fairchild 9300 AMD’s first second-source commercial product. 1971 Am3101 64-bit bipolar RAM Reverse-engineered from the Intel 3101 AMD’s first second-source RAM chip. 1971 Am7400 series TTL logic family (NAND Texas Instruments 7400 gates, inverters, flip-flops etc) series One of AMD’s biggest revenue drivers, especially military-grade versions. 1972-3 Am1488 Am1489 RS-232 line drivers & receivers National Semiconductor and Motorola Very successful in communications and computer peripherals. 1973 Am1702 2kib EPROM Intel 1702 One of the first memory products AMD second-sourced. 1975 Am9080 8-bit microprocessor Intel 8080 AMD’s first microprocessor (reverseengineered). It had faster variants than the Intel product. siliconchip.com.au Australia's electronics magazine November 2026  15 Fig.3: the die and wire bonding of AMD’s first RAM chip, the 64-bit Am3101, reverse-engineered from Intel’s 3101. Source: www.facebook. com/share/p/1ELjPTFSj7/ Fig.4: AMD’s first proprietary chip, the Am2501. Source: https:// overclockers.ru/blog/gummibear/ show/45878/pervaya-sereznayapobeda-sobstvennoj-arhitekturyamd-burevestnik-thunderbird Fig.5: the AM2501 in its package with the encapsulation material removed to reveal the die, bond wires and part of the lead frame. Source: Ken Shirriff (https://x.com/i/ status/1218982024070955009) Fig.6: a die photo of the Am2501 chip. Source: https://prohoster.info/ en/blog/novosti-interneta/amd-bylaosnovana-rovno-50-let-nazad-sostartovym-kapitalom-v-50-tysyach 16 Silicon Chip Intel did not sue because, at the time, patent and copyright laws did not apply to silicon layouts as long as they were not a direct copy based on the manufacturing masks (the blueprints to etch the silicon). In fact, Intel welcomed the Am3101 because, at the time, companies like Honeywell and Burroughs refused to design Intel chips into their systems unless a second-source supplier was available to guarantee supply and price competition. AMD thus also inadvertently helped Intel popularise its product. AMD soon built enough credibility and cash flow that it began moving beyond pure second-sourcing into original designs, most notably the Am2900 bit-slice processor family, introduced in 1975. Jerry Sanders often boasted during this period that while other companies competed on technology, AMD would compete on quality and service, and it worked. This second-source strategy was extremely successful and kept the young company alive during its most vulnerable years. The Am9300 The Am9300 was released in 1970. As a second-source component originally designed by Fairchild Semiconductor and designated 9300 (also second-sourced or cloned by National Semiconductor [DM9300] and Texas Instruments [SN9300]), it was a relatively simple but widely used logic device capable of parallel-load, shiftleft, shift-right and hold functions. It was a 4-bit medium-scale integration shift register containing around 85 transistors and 40 resistors, with 125 components in total and 26 logic gates. The Am9300 reflected AMD’s early strengths and market positioning. Although the company lacked the resources to design complex original chips in the beginning, its strength was in manufacturing high-quality bipolar logic devices with superior reliability and consistency. To do this, AMD applied its Fairchild-inherited process technology and rigorous quality control to produce versions that often outperformed the original source parts, particularly in military and industrial applications, where failure rates had to be extremely low. By successfully producing and selling the Am9300, AMD demonstrated Australia's electronics magazine its fundamental early capability: the ability to manufacture dependable, high-quality semiconductor parts and provide excellent customer support. The Am9300 was produced using bipolar TTL (transistor-transistor logic) process technology, which was the same high-speed bipolar process AMD inherited from its ex-Fairchild engineers. This bipolar TTL technology used gold doping for faster switching and was well-suited to the military, aerospace and industrial markets that AMD targeted. AMD’s self-designed products After AMD established its second-­ sourcing strategy, it had the confidence and cash flow to start developing its own chips, as shown in Table 2. Second-­sourcing remained AMD’s ‘bread and butter’ in the early years, and the revenue from that also allowed them to start designing its own chips. The Am2501 (1970) was AMD’s first proprietary/in-house designed logic chip, while the Am2900 family (1975) was its first in-house processor and by far its most important early innovation. We will now look at both of these chips and their significance. The Am2501 The Am2501 (Figs.4 & 5), was AMD’s first in-house designed product. It was a reflection of the company’s capabilities and strategic thinking in its earliest days. As a medium-scale integration TTL 4-bit binary up-only counter, it was a relatively straightforward but practical logic device that reflected AMD’s core strengths at the time, such as its bipolar process technology and the company’s focus on reliability and quality. Rather than attempting a highly complex chip as its first original design, AMD chose a useful, mid-­ complexity device that allowed it to prove its design and manufacturing abilities while minimising risk. Engineering schematics are not available, but it is estimated to have had around 150-250 transistors. Putting that into perspective, just one year later the world’s first microprocessor, the Intel 4004, had 2300 transistors, and today’s AMD processors have 8-30 billion transistors. The Am2501 also aligned with AMD’s early emphasis on serving military and industrial customers, markets that demanded high-reliability parts siliconchip.com.au and were willing to pay a premium for better quality and consistency. It was designed to meet rigorous US Military Standard, MIL-STD-883. Its commercial success helped validate AMD’s and Sanders’ strategy of competing on quality and customer service rather than cutting-edge complexity. It, along with the second in-house designed product, the Am2505, helped propel AMD’s annual sales to US$4.6 million by the end of 1971. The Am2900 The Am2900 family of chips represented a big leap forward for AMD, moving from designing relatively simple logic chips like counters and multipliers to creating an advanced, flexible microprocessor system that customers could use to build custom high-­performance multi-chip CPUs. The AM2900 series 4-bit bipolar bitslice microprocessor family, released in 1975, marked the transition of AMD from mostly second-sourcing to being a force in complex, original microprocessor products. The Am2900 family was generally considered better-performing than competing solutions when it was introduced in 1975. It had a higher speed, since the Am2901 (the core 4-bit ALU slice – see Fig.7) was faster than Intel’s own 3000-series bit-slice chips. AMD’s bipolar process technology gave it excellent speed for the era. Table 2: AMD’s first in-house chip designs Rank Year(s) Product(s) Type Status First 1970 Am2501 4-bit synchronous uponly binary counter First in-house designed product Second 1971 Am2505 4-bit by 2-bit two’scomplement digital multiplier Second in-house design; fastest digital multiplier on the market ... 1971-74 Various Smaller logic and memory chips More in-house designs Major milestone 1975 Am2900 family Bit-slice microprocessor architecture First major complex original architecture It also offered a very flexible, microprogrammable architecture. Designers could easily create high-performance custom CPUs of any bit width (8, 16, 32, 64 bits etc), which was very attractive for minicomputer makers, graphics systems and military applications, where cost was not such a major concern. In addition, it had superior instruction throughput; systems built with the Am2900 family often achieved significantly higher performance than systems based on the Intel 8080 or 8085. The Am2900 was an excellent choice for high-end applications, and because of its speed and flexibility, the Am2900 became very popular in highend minicomputers, disk controllers, graphics processors and military and aerospace systems. Fig.7: a die photo of the Am2901 with the metal layers included. Source: www.righto. com/2020/04/ inside-am2901amds-1970s-bitslice.html siliconchip.com.au Australia's electronics magazine Some examples of its use included the Atari arcade video games Tempest and Battlezone (which used four Am2901s for vector calculations), the Xerox Star Workstation, which featured an early GUI (graphical user interface) and a mouse, and the Chyron IV character generator used in television broadcasting, which was released in 1977. While the Intel 8080 (and AMD’s own Am9080 clone) was a good general-­purpose 8-bit microprocessor, the Am2900 was aimed at a higher performance tier. It was not a direct competitor to the 8080 because it was in a different class (bit-slice), designed for customers who needed more speed and customisation than a standard microprocessor could offer. The Am2900 family was widely regarded as one of the best-­performing bit-slice solutions available in the midto-late 1970s, giving AMD its first real reputation as a company that could deliver high-performance, original designs. The Am2900 family reflected AMD’s strengths at the time. Having built its early reputation on high-­ quality bipolar manufacturing and logic design (as demonstrated by chips like the Am2501 and Am2505), AMD expanded these capabilities to create a modular, high-performance bit-slice microprocessor architecture. With its bipolar process technology, AMD could deliver superior speed compared to competing MOS-based solutions. This leveraged AMD’s manufacturing expertise in high-speed bipolar logic and its increasing abilities in designing modular building-­ block chips. The Am2900 family thus represented AMD’s successful transition from a mostly second-source supplier to an innovative designer. November 2026  17 The Intel agreement As noted in Table 1, the Am9080 (Figs.8 & 9) was released in 1975. While based on the Intel 8080, it was an unauthorised, unlicensed reverse-engineered product. As was common in the industry at the time, AMD decapped an Intel processor, photographed the die layers and analysed its circuitry. AMD’s engineers then created their own masks and layout to manufacture a chip that was functionally identical and pin-compatible with the Intel 8080. It used the same instruction set and behaved the same way electrically. Both the AMD and Intel devices had 4500-6000 transistors (depending on how they are counted). The performance of the Am9080 was equivalent to the Intel 8080, but many customers preferred the Am9080 over the Intel 8080 specifically because of AMD’s better quality control and customer support. Jerry Sanders often highlighted the point that AMD didn’t try to beat Intel on speed with the Am9080; they beat them on quality and service. It should be noted, however, that AMD’s proprietary N-channel MOS manufacturing process allowed them to produce faster variants than Intel, with a maximum clock speed of 4MHz compared to Intel’s maximum speed of 3.125MHz. In 1976, they reached a formal second-­ source licensing agreement with Intel that legalised its production of the 8080 family, including support chips. This cleared any possible past intellectual property infringement issues (but others were to come later). This 1976 agreement was a major milestone for the company, as it gave them credibility and access to important markets, especially military & industrial, that often required approved second sources. AMD paid Intel an initial fee of US$25,000 plus US$75,000 per year in royalties for the license. The Am2900 and Am9080 You may have noticed that both the Am2900 and Am9080 were released to market at the same time, in 1975, the Am2900 being AMD’s own design and the Am9080 based on Intel’s 8080. This marked AMD’s ‘dual entry’ into the microprocessor market. There was no functional overlap, as the two products served different purposes. The Am9080 was suitable for mass-market microcomputing and early PCs (like some S100 bus systems), industrial controllers and hobbyists, where its low cost and monolithic design were prioritised over speed. The Am2900 was for high-end enterprise systems, like the minicomputer and mainframe markets supplied by Digital Equipment Corporation (DEC) and Data General. They required customised word lengths and high speeds for scientific applications and graphics terminals, with little regard to cost or the complexity of building a modular processor from multiple chips. To give an idea of prices, the Am9080 cost about US$0.50 per chip to manufacture and sold for US$1020, undercutting Intel, although the military versions sold for US$700. It should be noted that AMD did not focus on the low-cost end of the Am9080 market because they could sell them to the military for US$700, while still only spending US$0.50 per unit on manufacturing. The Am2900 cost hundreds of dollars for a full family of chips, including around US$15-30 for individual Am2901 ALUs, making it cost-­ prohibitive for ordinary consumers. Second-source success to innovation (1978–1985) The combination of the successful Intel 8080 licensing agreement in 1976 and the strong market reception of the Am2900 bit-slice processor family gave AMD its first taste of prosperity. By the late 1970s, the company had grown rapidly, built a solid reputation for quality, and established itself as a major player in the military, industrial and early personal computer markets. However, this success also created a strategic dilemma. As the personal computer revolution accelerated in the early 1980s, it became clear that the future of the microprocessor industry would be dominated by general-­ purpose CPUs rather than modular bit-slice designs. AMD’s significant dependence on second-sourcing and logic chips left it vulnerable. The 1982 Intel agreement In 1982, AMD signed a 10-year technology exchange agreement with Intel that was heavily driven by IBM, who wanted a reliable second-source supplier before it would commit to using the Intel 8086 and 8088 in the IBM PC (first released August 1981). Under this agreement, Intel gave AMD complete blueprints and microcode to officially manufacture exact clones of these products. This arrangement further enhanced AMD’s recognition as a major silicon manufacturer. Fig.8: a die photo of the AMD Am9080A microprocessor (AM9080ADIB). The layout was AMD’s own, as they reverse-engineered the Intel 8080. Source: https://w.wiki/RGet Fig.9: the Am9080 reverse-engineered clone of the Intel 8080. Source: https://w.wiki/RGf2 18 Silicon Chip Australia's electronics magazine siliconchip.com.au Fig.10: the AMD 8086 clone came in a standard dual-inline package that was compatible with Intel’s. Source: https://w. wiki/Q8N There is an obvious timing discrepancy. IBM released the PC before Intel had a second-source supplier. IBM started production of the PC on the basis that Intel had promised to find a second-source supplier, and the preliminary deal with AMD was agreed to in October 1981, with a formal contract signed in February 1982. The agreement provided for Intel to provide the layouts for the 8086 (Fig.10), 8088, 80186 and 80286 processors. AMD ‘paid’ Intel by developing peripheral chips or other technology of equivalent complexity. Expiry of the 1976 agreement postponed The 1982 agreement mentioned above did not replace the former 1976 agreement, but added to it. While the original 1976 cross-­ l icensing agreement with Intel for the 8080/ Am9080 was to expire in 1985, the new agreement effectively extended the technology-­ sharing arrangement to December 1995. The 1982 agreement was effectively a barter system, with no payments or royalties involved; exchanges were based on technical complexity, according to a formula such as a point value assigned to each die based on die size, transistor count and engineering hours. To ‘earn’ the blueprints for a new Intel CPU, AMD had to trade back alternative chips like memory controllers, networking chips, mathematics coprocessors etc that equalled the same number of complexity points as Intel was trading. Intel backs out these designs, claiming they were “substandard” and not of sufficient value to trade for the 80386. They thus claimed AMD had no right to manufacture the chip. Intel did not cancel the contract; they just dragged things out. Eventually, AMD realised Intel was never going to give them the blueprints for the 80386 and filed for legal arbitration in April 1987. The case dragged on for nearly five years. During those years, Intel had a 100% monopoly on the 80386 market. This refusal forced AMD into a difficult situation. The company could no longer rely on easy access to Intel’s latest designs and now faced a stark choice. It had to either remain a successful but secondary supplier of older-­generation parts, or take the enormous risk of developing its own competitive x86 processors. AMD’s leadership, still led by the flamboyant Jerry Sanders, chose the more ambitious but high-risk path. The company began investing heavily in original CPU design talent and advanced manufacturing processes. This period marked the beginning of AMD’s painful but necessary evolution from a clever second-source company into a true innovator. The experience gained during the Am2900 years, particularly in complex logic design, bipolar process technology and high-reliability manufacturing provided a strong technical foundation. However, moving into direct competition with Intel on general-purpose microprocessors would require entirely new skills, massive investment and a willingness to challenge Intel both technically and legally. By the mid-1980s, AMD had outgrown its original business model (second-­sourcing). The stage was now set for one of the most dramatic events in semiconductor history. That is, AMD’s attempt to break Intel’s monopoly on the 32-bit 80386 and 80486 architectures. Breaking Intel’s monopoly By the mid-1980s, Intel had successfully turned the x86 architecture into a near-monopoly. With its refusal to grant AMD a license to produce the 80386 microprocessor, AMD faced a critical threat. Undeterred, AMD embarked on one of the most ambitious and legally risky projects in its history. It decided to reverse-engineer the Intel 80386. The company assembled two separate teams. One team, which had never seen Intel’s 386 documentation or microcode, analysed the physical chip through decapping, microscopic photography and electrical testing to create detailed functional specifications. A second, completely isolated team designed a new chip layout based solely on those specifications. This ‘cleanroom’ technique was designed to avoid any direct copying of Intel’s copyrighted layout material, while achieving fully functional compatibility. They did, however, copy the microcode, the software that controls the chip. There are some variations of exactly what happened, but we believe this is the correct version. Fig.11: an Am386 processor in a TQFP package. It was the first non-Intel manufactured 80386 processor. Source: https://w.wiki/RGev When Intel introduced the innovative 32-bit 80386 in 1985, it realised it was going to be hugely popular and a cash cow. They did not want to share this with AMD. Intel’s plan was to keep AMD away from the 80386 or subsequent processors. AMD offered Intel dozens of its original peripheral chip designs to pay for the 80386, but Intel rejected siliconchip.com.au Australia's electronics magazine November 2026  19 In 1991, AMD released the Am386, the first non-Intel 80386 processor (Fig.11). The chip was pin-­compatible with Intel’s design and available in both 25MHz and 40MHz versions. Technically, the Am386 was a great success. It delivered performance comparable to, and in some cases, slightly better than Intel’s own 80386 (released in 1985) due to minor architectural optimisations and AMD’s manufacturing strengths. More importantly, it was significantly cheaper and, for the first time, offered customers a real alternative in the 32-bit x86 market. Intel immediately sued AMD for copyright infringement, focusing primarily on the microcode. The resulting legal battle became one of the most important cases in Silicon Valley history. In 1992, a US federal jury ruled in AMD’s favour on the important microcode copyright claim, determining that AMD had not illegally copied Intel’s code and that they were allowed to do it under the 1976 cross-licensing agreement and the 1982 second-­sourcing contract. The case eventually reached the US Supreme Court, which declined to hear Intel’s appeal, effectively upholding AMD’s victory. This was an extremely important legal win, as it established that functional compatibility through clean-room reverse engineering was legally defensible. It fundamentally weakened Intel’s ability to maintain control over the x86 architecture through copyright claims. The curious copyright symbol One of the most curious details of the Am386 story was that AMD Am386 processors physically carried an Intel copyright symbol (“© Intel 85”) etched onto the silicon die – see Fig.12. Fig.13: an AMD 80486DX2 die. The apparently solid or hatched rectangular are likely cache memories. Source: https://w.wiki/ RGez Fig.14: an early Am486 chip (40MHz version) in a PGA (pin grid array) package; a package style AMD stuck with until the early 2020s. Source: https://w.wiki/RGey AMD incorporated the Intel copyright symbol to acknowledge that it had used Intel’s microcode (or at least that the chip was based on elements of the 80386; once again, there are different versions of the precise details). All Am386s shipped with this notice. Reverse-engineering the 80486 After the success and legal victory surrounding the Am386, AMD moved aggressively into the 80486 market. However, this proved far more difficult. The 80486 contained significantly more complex microcode and integrated features (such as an on-chip floating-point unit in later versions). AMD’s first 486-compatible processors, the Am486 series, launched in 1993 (see Figs.13 & 14). They were considered respectable but generally lagged somewhat behind Intel’s products in performance. However, they gave customers a choice and kept downward pressure on Intel’s pricing. AMD made a major agreement with Compaq to supply the Am486 for the Fig.12: the “© Intel 85” notice in the silicon of an Am386 processor die. Source: https://w. wiki/RGeu budget Presario home PCs, which enabled them to undercut the price of other computers using the Intel 80486. The ‘386/486 wars’, as the CPU competition between AMD and Intel became known, fundamentally changed AMD’s path. The company transformed from a respected secondary supplier into a legitimate challenger to Intel’s dominance. More importantly, these ‘wars’ forced AMD to invest heavily in engineering talent and manufacturing capabilities. The legal and technical experience gained during this period would prove essential years later when AMD developed its own in-house designed K5, K6, K7 and eventually the groundbreaking Zen architecture, which we will cover in a later part of this series. In summary, by the mid-1990s, AMD had proven it could compete at the highest level of the microprocessor industry, not just by copying Intel, but by outmanoeuvring and out-­ competing them both legally and in the marketplace. Thus, the days of being a pure second-source company were over. AMD’s next move was to design its own processors and make the strategic decision to cease being a second-­ source supplier. Next month In part two of this series next month, we will pick up where we left off, in the mid-1990s, after the launch of the AM486. This was the start of a new era for AMD – no longer merely content to clone or license Intel processors, they started designing their own chips to SC compete directly with Intel’s. 20 Silicon Chip Australia's electronics magazine siliconchip.com.au Table 3: AMD’s main processor families Second-source, cloning and reverse engineering era (1969-1995) 1975 Am9080 and Am2900 The Am9080 was a reverse-engineered clone of the Intel 8080, while the Am2900 was an in-house designed bit-slice processor family. 1982 Am8086 A licensed version of Intel’s 8086 processor. 1984 Am286 A licensed version of Intel’s 80286 processor with higher clock speeds. 1988 Am29000 The first of a family of 32-bit RISC processors used in embedded applications like laser printers, parts of which were incorporated into the later K5 processor. 1991 Am386 A reverse-engineered 80386 clone that triggered a legal battle with Intel over microcode copyright. It was popular due to its 40MHz speed, beating Intel’s 33MHz. 1993 Am486 A high-performance reverse-engineered 80486 clone that culminated in a longterm deal with Compaq. Independent architecture and the GHz race (1996-2003) 1996 AMD K5 AMD’s first independently designed x86 processor. It was a little slower than Intel’s Pentium overall. It was based on elements of an unreleased version of the Am29000. 1997 AMD K6 The K6 integrated multimedia instructions (MMX) and drastically drove down the cost of desktop personal computers. 1999 AMD K7 Athlon The debut of the Athlon brand with a drastically superior architecture. It became the fastest x86 processor on the consumer market, significantly outperforming the Pentium III at the same clock speed. 2000 Athlon 1000 AMD broke the 1GHz barrier before Intel. Innovation leadership (2003-2007) 2003 Athlon 64 and Opteron (K8) AMD introduced the x86-64 64-bit system that’s still in use today. It entered the server market with the Opteron, while the Athlon 64 dominated gaming PCs with an on-die memory controller. 2005-9 Athlon 64 X2 The first native dual-core x86 processor. 2007 Phenom and Phenom II (K10) Native quad-core architectures. AMD also released K10 Opteron budgetfriendly server processors. Dark Years and the Bulldozer architecture (2008-2016) 2011 Bulldozer and FX Series The FX series used ‘modules’, sharing resources between integer cores. It suffered from poor single-threaded efficiency as a result and had high power consumption. 2011 The first APUs (accelerated processing units) AMD introduced the APU under the Fusion project, merging a CPU and a powerful Radeon graphics die for budget computing. The Zen Renaissance and the Modern Era (2017-Present) 2017 Zen 1 (Ryzen 1000 series) Zen 1 introduced simultaneous multi-threading (SMT) to AMD processors and brought 8-core/16-thread power to consumers on a 14nm process. 2018 Zen+ (Ryzen 2000 series) A process shrink (12nm) that improved clock speeds, latency and power efficiency. 2019 Zen 2 (Ryzen 3000 series) The first consumer CPU using a multi-die chiplet design, vastly lowering manufacturing costs and exploding core counts on 7nm. 2020 Zen 3 (Ryzen 5000 series) Unified 8-core complex (CCX) that eliminated internal latency. Introduced 3D V-Cache variants for unmatched gaming performance. 2022 Zen 4 (Ryzen 7000 series) Moved to the modern AM5 LGA socket, supporting DDR5, PCIe 5.0, and AVX512 instructions on a 5nm process. Zen 4/4c EPYC server processors with 98/128 cores became available. 2024 Zen 5 (Ryzen 9000 series) Delivers large instructions-per-cycle (IPC) gains and specialised hardware execution branches for artificial intelligence on a 4nm process. Zen 5c server chips have up to 192 cores! est. 2027 Zen 6 (Next-Gen platform.) Continues to push chiplet densities on the extended AM5 ecosystem. Year(s) Comments Processor siliconchip.com.au Australia's electronics magazine November 2026  21