Inventing The Microprocessor: The Intel 4004

We recently looked at the origins of the integrated circuit (IC) and the calculator, which was the IC’s first killer app, but a surprise twist is that the calculator played a big part in the invention of the next world-changing marvel, the microprocessor.

There is some dispute as to which company invented the microprocessor, and we’ll talk about that further down. But who invented the first commercially available microprocessor? That honor goes to Intel for the 4004.

Path To The 4004

Busicom calculator motherboard based on 4004 (center) and the calculator (right)
Busicom calculator motherboard based on 4004 (center) and the calculator (right)

We pick up the tale with Robert Noyce, who had co-invented the IC while at Fairchild Semiconductor. In July 1968 he left Fairchild to co-found Intel for the purpose of manufacturing semiconductor memory chips.

While Intel was still a new startup living off of their initial $ 3 million in financing, and before they had a semiconductor memory product, as many start-ups do to survive they took on custom work. In April 1969, Japanese company Busicom hired them to do LSI (Large-Scale Integration) work for a family of calculators.

Busicom’s design, consisting of twelve interlinked chips, was considered a complicated one. For example, it included shift-register memory, a serial type of memory which complicates the control logic. It also used Binary Coded Decimal (BCD) arithmetic. Marcian Edward Hoff Jr — known as “Ted”, head of the Intel’s Application Research Department, felt that the design was even more complicated than a general purpose computer like the PDP-8, which had a fairly simple architecture. He felt they may not be able to meet the cost targets and so Noyce gave Hoff the go-ahead to look for ways to simplify it.

Hoff realized that one major simplification would be to replace hard-wired logic with software. He also knew that scanning a shift register would take around 100 microseconds whereas the equivalent with DRAM would take one or two microseconds. In October 1969, Hoff came up with a formal proposal for a 4-bit machine which was agreed to by Busicom.

This became the MCS-4 (Micro Computer System) project. Hoff and Stanley Mazor, also of Intel, and with help from Busicom’s Masatoshi Shima, came up with the architecture for the MCS-4 4-bit chipset which consisted of four chips:

  • 4001: 2048-bit ROM with a 4-bit programmable I/O port
  • 4002: 320-bit DRAM with 4-bit output port
  • 4003: I/O expansion that was a 10-bit static, serial-in, serial-out and parallel-out shift register
  • 4004: 4-bit CPU

Making The 4004 Et Al

In April 1970, Noyce hired Federico Faggin from Fairchild in order to do the chip design. At that time the block diagram and basic specification were done and included the CPU architecture and instruction set. However, the chip’s logic design and layout were supposed to have started in October 1969 and samples for all four chips were due by July 1970. But by April, that work had yet to begin. To make matters worse, the day after Faggin started work at Intel, Shima arrived from Japan to check the non-existent chip design of the 4004. Busicom was understandably upset but Faggin came up with a new schedule which would result in chip samples by December 1970.

Faggin then proceeded to work 80 hour weeks to make up for lost time. Shima stayed on to help as an engineer until Intel could hire one to take his place.

4004 architecture
4004 architecture by Appaloosa CC BY-SA 3.0

Keeping to the schedule, the 4001 ROM was ready in October and worked the first time. The 4002 DRAM had a few simple mistakes, and 4003 I/O chip also worked the first time. The first wafers for the 4004 were ready in December, but when tried, they failed to do anything. It turned out that the masking layer for the buried contacts had been left out of the processing, resulting in around 30% of the gates floating. New wafers in January 1971 passed all tests which Faggin threw at it. A few minor mistakes were later found and in March 1971 the 4004 was fully functional.

In the meantime, in October 1970, Shima was able to return to Japan where he began work on the firmware for Busicom’s calculator, which was to be loaded into the 4001 ROM chip. By the end of March 1971, Busicom had a fully working engineering prototype for their calculator. The first commercial sale was made at that time to Busicom.

The Software Problem

Now that Intel had a microprocessor, they needed someone to write software. At the time, programmers saw prestige in working with a big computer. It was difficult enticing them to stay and work on a small microprocessor. One solution was to trade hardware, a sim board for example, to colleges in exchange for writing some support software. However, once the media started hyping the microprocessor, the college students came banging on Intel’s door.

To Sell Or Not To Sell

Intel D4004
Intel D4004 by Thomas Nguyen CC BY-SA 4.0

Intel’s market was big computer companies and there was concern within Intel that computer companies would see Intel as a competitor instead of a supplier of memory chips. There was also a question about how they would support the product. Some at Intel also wondered whether or not the 4004 could be used for more than just a calculator. But at one point Faggin used the 4004 itself to make a tester for the 4004, proving that there were more uses.

At the same time, cheap $ 150 handheld calculators were creating difficulties for Busicom’s more expensive $ 1000 desktop ones. They could no longer pay Intel the agreed contract price. But Busicom had exclusive rights to the MCS-4 chips. And so a fateful deal was made wherein Busicom would pay a lower price and Intel would have exclusive rights. The decision was made to sell it and a public announcement was made in November 1971.

By September 1972 you could buy a 4004 for $ 60 in quantities of 1 to 24. Overall, around a million were produced. To name just a few applications, it was used in: pinball machines, traffic light controllers, cash registers, bank teller terminals, blood analyzers, and gas station monitors.

Contenders For The Title

Most inventions come about when the circumstances are right. This usually means the inventors weren’t the only ones who thought of it or who were working on it.

AL1 as a microprocessor
AL1 as a microprocessor by Lee Boysel

In October 1968, Lee Boysel and a few others left Fairchild Semiconductor to form Four-Phase Systems for the purpose of making computers. They showed their system at the Fall Joint Computer Conference in November 1970 and had four of them in use by customers by June 1971.

Their microprocessor, the AL1, was 8-bit, had eight registers and an arithmetic logic unit (ALU). However, instead of using it as a standalone microprocessor, they used it along with two other AL1s to make up a single 24-bit CPU. They weren’t using the AL1 as a microprocessor, they weren’t selling it as such, nor did they refer to it as a microprocessor. But as part of a 1990 patent dispute between Texas Instruments and another claimant, Lee Boysel assembled a system with an 8-bit AL1 as the sole microprocessor proving that it could work.

Garrett AiResearch developed the MP944 which was completed in 1970 for use in the F-14 Tomcat fighter jet. It also didn’t quite fit the mold. The MP944 used multiple chips working together to perform as a microprocessor.

On September 17, 1971, Texas Instruments entered the scene by issuing a press release for the TMS1802NC calculator-on-a-chip, with a basic chip design designation of TMS0100. However, this could implement features only for 4-function calculators. They did also file a patent for the microprocessor in August 1971 and were granted US patent 3,757,306 Computing systems cpu in 1973.

Another company that contracted LSI work from Intel was the Computer Terminal Corporation (CTC) in 1970 for $ 50,000. This was to make a single-chip CPU for their Datapoint 2200 terminal. Intel came up with the 1201. Texas Instruments was hired as a second supplier and made samples of the 1201 but they were buggy.

Intel’s efforts continued but there were delays and as a result, the Datapoint 2200 shipped with discrete TTL logic instead. After a redesign by Intel, the 1201 was delivered to CTC in 1971 but by then CTC had moved on. They instead signed over all intellectual property rights to Intel in lieu of paying the $ 50,000. You’ve certainly heard of the 1201: it was renamed the 8008 but that’s another story.

Do you think the 4004 is ancient history? Not on Hackaday. After [Frank Buss] bought one on eBay he mounted it on a board and put together a 4001 ROM emulator to make use of it.

[Main image source: Intel C4004 by Thomas Nguyen CC BY-SA 4.0]