A Brief History of the MOS transistor, Part 6: Intel

May 08, 2023

Robert Noyce and Gordon Moore are inextricably linked to the three most important commercial semiconductor companies in the history of semiconductors. First William Shockley brought Noyce and Moore together at his Shockley Semiconductor Laboratory in Palo Alto, California to develop silicon transistors. Although the company failed to develop commercially successful products, the team that William Shockley assembled then founded Fairchild Semiconductor, perhaps the most important semiconductor company ever created because it spawned so many other semiconductor companies and was a major factor in the creation of Silicon Valley.

In September 1957, eight Shockley employees – Robert Noyce, Gordon Moore, Julius Blank, Victor Grinich, Jean Hoerni, Eugene Kleiner, Jay Last, and Sheldon Roberts – became fed up with Shockley's autocratic style, resigned en masse, and founded Fairchild as a subsidiary of Fairchild Camera and Instrument, an East Coast technology company with photographic and aircraft manufacturing operations. In short order, Fairchild Semiconductor developed the planar process, the semiconductor industry's bedrock technological and manufacturing foundation for more than half a century, and the integrated circuit, based on the planar process. Fairchild's transistors and ICs took center stage in the electronics market.

By 1968, Noyce and Moore saw that Fairchild was crumbling. The company had become dominated by east coast management style, and it seemed that Fairchild was spawning competing semiconductor companies almost weekly. There was a serious brain drain taking place. On top of that, Noyce had been passed over when it came time to name a CEO for Fairchild, in favor of C. Lester Hogan. Seeing a sinking ship, Noyce and Moore left Fairchild in 1968 and founded NM Electronics, which quickly became Intel.

Noyce and Moore founded Intel to be a memory IC company because Moore determined that semiconductor memory would most likely benefit from the exponential scaling trend that he’d described in a 1965 article published in "Electronics" magazine. That trend became known universally as Moore's Law, which is not actually a law. It's a self-fulfilling prophecy. Semiconductor companies soon learned that they either followed Moore's Law, or they died.

Because of Fairchild's strong history in bipolar semiconductors and because his Fairchild R&D lab had extensively studied MOS ICs, Moore remained agnostic about bipolar versus MOS ICs. He was willing to try making IC memories with both types of process technologies.

Intel's first memory products were the 64-bit 6101 bipolar RAM and the 1-kbit 3301 bipolar ROM. These were not large devices, but they proved that Intel had built a working semiconductor fab by 1969. Next out the door was the 1101, a 256-bit PMOS SRAM. The 1101 proved to be too slow and too small for mainframe manufacturers to use as the main memory in their computers, so it was not commercially successful, but it did use the silicon-gate process developed by Federico Faggin at Fairchild, just before Noyce and Moore left that company. Faggin still worked for Fairchild at this point, but he and his silicon-gate technology would become crucial in Intel's immediate future.

Meanwhile, Intel needed a big seller, and its next memory IC, the revolutionary 1-Kbit 1103 DRAM, would prove to be a hit when it was introduced in October 1970, and, after several iterations of the die layout, it worked. The 1103 DRAM and its descendants quickly put magnetic core memory makers out of business. However, while Intel was attempting to develop a high-volume standard memory IC business, the company started taking custom IC contracts to bring in some revenue while trying to develop a commercially lucrative memory product.

One of the companies that approached Intel with a custom IC contract was a Japanese calculator company named Busicom, which had developed the design for a programmable desk calculator. Busicom had already signed an agreement with Mostek to develop a custom calculator chip for a simple 4-banger calculator that would eventually be named the Busicom Junior. That calculator chip was the Mostek MK6010. Single-chip calculator chips were the apex of MOS LSI design in the late 1960s, and many chip vendors developed them.

However, Busicom also planned to develop a high-performance, programmable calculator called the 141-PF. Busicom's Masatoshi Shima developed a modular chipset for the 141-PF calculator. His design proposed twelve different LSI chips including a 3-chip CPU, ROMs that would add optional functions, a keyboard controller chip, a printer controller chip, and a display controller chip. Instead of binary, the chipset used decimal arithmetic for calculations, which was the basis for Busicom's and other mechanical calculators.

Shima presented his chipset proposal to Intel in 1969. Each chip would require 3000 to 4000 transistors, with 36 to 40 pins per package. Intel's Marcian "Ted" Hoff evaluated Shima's proposal and concluded that Intel could not develop or manufacture the calculator chipset as Shima had designed it, for three reasons. First, Intel's recently developed silicon-gate MOS process was capable of cramming 2000 or so transistors onto a chip, roughly half the number that Shima's design would require. Second, Intel was a memory-chip company, so the only IC packages it had on hand were narrow, 16- and 18-pin ceramic DIPs. These packages could not accommodate Shima's design. Third, Intel didn't have enough chip designers to create all of Shima's different chip designs. There were only two such designers in the entire company.

Hoff was Intel's only system-level engineer. He was familiar with the Digital Equipment Corp (DEC) PDP-8 minicomputer and realized that he could reduce the logic complexity in Shima's design by shifting more of the calculations to software running on a general-purpose, binary CPU. However, a single-chip version of DEC's 12-bit PDP-8 minicomputer architecture was equally beyond Intel's integration capabilities at the time, so Hoff developed an architecture for a 4-bit processor, which more closely matched the computational needs of a calculator.

Shima saw promise in Hoff's proposal but was not overly impressed with it because the proposal consisted of a very high-level architectural description and lacked implementation details. Despite its unfamiliar, software-centric architecture, Busicom's execs eventually approved Hoff's design proposal because it appeared to offer a less costly solution.

Shima returned to Japan in 1970 to redevelop his calculator design based on Hoff's architecture. He returned to Intel in early 1971 and found that no progress had been made on developing the microprocessor chipset during the intervening months. Worse, when he arrived, he found that Ted Hoff was now working on a different project. Instead, Federico Faggin, who’d joined Intel just a week before Shima arrived from Japan, was now assigned to the project.

Here's how Faggin described the situation on his Web site:

"I was hired to be the project leader for the "Busicom project." In 1969, Busicom had developed the design for a family of calculators, and they wanted Intel to transfer their design into silicon. Their design was a special-purpose, CPU-based machine with macro-instructions, ROM, and shift-register read-write memory, partitioned in 7 different chips. The CPU function was distributed onto three separate chips. When Ted Hoff saw the Busicom design, he was surprised by its complexity and proposed a simplified architecture based on a more general-purpose, single-chip, 4-bit CPU, and separate chips for ROM with I/O, RAM with I/O, and shift register. After the basic architecture and specification of this four-chip set, called MCS-4, was completed, the project was transferred to the MOS Design group for implementation.

"For a typical custom design, the customer usually provided a verified logic design to the MOS vendor for implementation in silicon. The semiconductor company would then translate the customer's logic design into appropriate MOS circuits, do the chip layout, create the masks, the customized testing software, and finally produce the chips for the customer. Companies doing MOS custom chips over a number of years had developed a design methodology and a number of pre-characterized circuit building blocks to aid in error-free and rapid chip development.

"The original design that Busicom wanted Intel to translate into seven custom chips was accordingly already designed, and it was built and verified to be correct at the logic gate level. Hoff's proposal, however, was only conceptual and no logic design or verification had been done. Since Intel was dedicated to memory chip design, it had no experience with random logic chip design; it had no design methodology and no pre-characterized circuit building blocks.

"Furthermore, the style of design required with silicon gate technology was quite different than random logic with metal gate, for which methodologies already existed. The characterization and production testing required testers and testing methods that Intel also didn't have. The day I joined Intel, I had in front of me the task of performing the entire logic, circuit, and layout design of the four chips, characterizing and transferring them to production, in six months. That's the schedule Intel had agreed with Busicom six months earlier, but no work had been done since then! I also ended up designing and building the characterization and the wafer sorting testers as well, while the final-test tester was being purchased."

Shima assisted Faggin with all the development phases of the project. He learned how to design integrated circuits using the methodologies that Faggin developed for Intel. He assisted with checking the logic design, the chip layout, and the Rubylith artwork for the microprocessor chipset. He also developed most of the logic design of the 4004 CPU, under Faggin's supervision. Working at a furious pace, Faggin and Shima produced working samples of the Busicom calculator chipset in only nine months. Faggin brought the first 4004 microprocessor to life on a tester in January, 1971.

In the middle of 1971, Faggin learned that Busicom was in some financial trouble during a telephone conversation with Shima, who had returned to Japan by then. At the same time, Busicom was asking Intel for price concessions on the microprocessor chip set. This information helped Noyce make the new agreement with Busicom, trading the requested price concessions for the right to market the world's first commercial microprocessor chipset, the Intel MCS-4.

By the end of 1971, Intel was in both the semiconductor memory business and the microprocessor business. The company rode these technologies to become the world's top-selling semiconductor maker. At first, Intel developed bipolar and PMOS chips. The Intel 1103 DRAM and the 4-bit 4004 and 8-bit 8008 microprocessor (introduced in 1972) were PMOS devices. When the company realized that only MOS process technologies would be able to ride Moore's Law, Intel dropped its bipolar process. After IBM proved that NMOS could be made, Intel switched to NMOS so that it could make faster devices. The 8-bit Intel 8080 (introduced in 1974) and 8085 (introduced in 1976) microprocessors were NMOS devices. After Hitachi proved that CMOS could deliver NMOS speeds with lower power consumption, Intel switched to CMOS for its microprocessor designs. Over the years, Intel has been willing to ride the best available process horse, whether that's bipolar or some flavor of MOS, and that ride's now lasted more than 50 years.

References

To the Digital Age: Research Labs, Start-Up Companies, and the Rise of MOS Technology, Ross Knox Bassett, 2002

Silicon: From the Invention of the Microprocessor to the New Science of Consciousness, Federico Faggin, 2021

The Intel 4004 Microprocessor and the Silicon Gate Technology, Federico Faggin's Web site

Say Happy 50th Birthday to the Microprocessor, Part I, Steve Leibson

Say Happy 50th Birthday to the Microprocessor, Part 2, Steve Leibson