The birth of the microprocessor: the Intel 4004
The 4004 was the first commercially produced microprocessor, operating at up to 740 kHz, with single-word instructions requiring 10.8 μs and two-word instructions requiring 21.6 μs. Let’s take a closer look.
After watching men land on the moon, seeing the future depicted in animated series such as The Jetsons, and switching from tube radios to transistor radios, there seemed to be only one remarkable invention: the microprocessor. Newly founded and ready to ride the wave of the nascent semiconductor industry, the task fell to Intel. However, the accolade could easily have been won by any of the many other players in the industry had it not been for the dedication and skill of Federico Faggin. Let’s look at the history of the 4004 (Figure 1).
In the 1960s, computing meant mainframe computers, such as IBM’s System/360, with available processing performance shared among many users and programs. Memory was implemented using magnetic ferrite cores, and available integrated circuits (ICs) were mostly fabricated using transistor-transistor logic (TTL). Intel was convinced that the future lay in metal-oxide-semiconductor (MOS) technology to supplant magnetic core memories.
In 1969, the Japanese company Busicom asked Intel to develop a collection of integrated circuits to form the basis of their first electronic calculator, the prototype 141-PF printing calculator. [1 and 2]. Their approach used 12 chips, three of which formed the central processing unit (CPU). The goal was to develop a platform for electronic calculators that could be modified simply by changing the read-only memory (ROM) integrated circuits. Intel’s team saw room to optimize the design, in part because the required packages didn’t match their 16-pin standard in use at the time. A four-chip solution was offered with the processor integrated into a single package. The “Busicom Project” was thus born and, after some design work undertaken, the project lay dormant towards the end of the year.
TTL was the technology of choice at the time, with a range of logic devices and memories available. MOS showed promise, being able to deliver 10 times the density of TTL for the same cost and power dissipation. However, his speed was much slower and he was generally considered unreliable. Intel had successfully brought a MOS static random access memory (SRAM) to market, the 1101, but initial production could only produce two working chips per wafer. . The speed issue was related to the capacity of the door caused by a combination of the use of aluminum and the manufacturing process. It had been determined that this capacitance could be reduced by using a silicon gate, allowing the gate to be “self-aligned” and the entire structure to be smaller. Unfortunately, the amorphous silicon approach used to date has proven to be brittle and prone to cracking during oxide steps.
Federico Faggin joined Intel in the early 1970s after working on MOS technology at Fairchild. There he had solved a series of critical problems to make the design of a microprocessor possible. First, using vapor deposition, he had replaced the amorphous silicon gate with polycrystalline silicon, leading to a much more reliable manufacturing process. Second, he had developed a method for buried contacts that removed the need for aluminum, resulting in denser circuitry. And, finally, he had created a bootstrap loading approach that allowed full and faster output signal oscillations than before. Applying the technology to Fairchild’s 3708, an eight-channel analog multiplexer, resulted in a design four times faster, with 10 times lower leakage current and multiplexing transistors two and a half times smaller than the 3705, its predecessor which used metal-gate MOS fabrication.
With Intel’s designers busy working on the 1103, their first dynamic MOS RAM (DRAM), Faggin received what little documentation there was from the Busicom project and struck out on his own. The four chips consisted of the 4001, a 2048-bit metal-masked ROM; the 4002, a 320-bit RAM; the 4003, a 10-bit static shift register with serial I/O and closed parallel outputs; and the 4004 , a 4-bit processor whose design was incomplete. The 4000 product numbering used departs from Intel’s standard approach. Still, this allowed the four chips to be seen as compatible parts of a complete microcomputer platform that later became known as MCS-4 (Micro-Computer Set) (Figure 2).
The sudden arrival of Masatoshi Shima, Busicom’s lead engineer, just days after Faggin began his new role, caused problems when Shima learned that the chips his calculator depended on hadn’t even been designed. Under pressure to meet Busicom’s schedule, Faggin agreed to deliver a fully operational chipset by December 1970. The challenges were immense. Each device had to be designed and laid out by hand while applying the latest MOS technology during the manufacturing process. The Intel team was used to having memories that were repetitive in functionality, so the 4004 pushed everyone to the limit and tested the limits of MOS manufacturing.
Shima was given permission to stay in the United States for six months to support Faggin during development. As the project progressed, more designers and layout artists were assigned. Around October, the first samples of the 4001 arrived, followed closely by the 4003. Both devices worked for the first time and even met the design goals set for clock speed and signal and power margins. food. The 4002 only needed a minor fix which was easily fixed.
Shortly before New Year’s Eve, 4004 was delivered. With the test probe carefully lowered into place, the first array did not work. Others died, and then another wafer also showed no signs of life. Close inspection of the wafer using a microscope showed that the buried contact layer was completely missing, leaving most of the transistor gates unconnected. Delivery of a new batch of inserts three weeks later showed that the design was fully functional.
The 4004 was the first commercially produced microprocessor, operating at up to 740 kHz, with single-word instructions requiring 10.8 µs and two-word instructions requiring 21.6 µs (picture 3). With only 16 pins available, two were dedicated to a bi-phase clock, two to power, and another two to reset and test. Four were assigned to a 4-bit bidirectional data bus, one for a ROM select signal and another for a synchronization output. The remaining four pins were used to select RAM banks. The processor had no support for interrupts but had a three-level stack. 4001 ROM chips included a 4-bit I/O port, while 4002 RAM devices had a 4-bit output port. The 4003 shift register was designed to interface with these ports. Together, these interfaces provided a method for interfacing with keyboards, printers, and other peripherals.
The 4004 reaches a wider audience
The MCS-4 design was exclusive to Busicom, meaning it could not be sold to anyone else. This proved frustrating for Faggin, who saw the incredible potential offered by this freely programmable processing platform. To prove its potential beyond electronic calculators to Intel’s management, Faggin built a wafer sort tester for the 4004 using the MCS-4 chipset. Busicom was also in financial trouble so when Intel offered them the chips at a lower price in exchange for canceling the exclusivity deal, they accepted. And, with that, the MCS-4 and 4004 became available to everyone.
Busicom used the MCS-4 chipset to create other applications, such as ATMs, cash registers, and billing machines. Other companies used it to power the first “Flicker” microprocessor-controlled pinball game, a bowling simulator, electric typewriters and even an early voting system. Faggin continued designing the processor, taking over Intel’s 8-bit Project 1201, which became the 8008 processor and launched in 1972. Both processors received upgrades, providing engineers with the 4040 and 8080 in 1974. At that time, Faggin believed that Intel was only interested in microprocessors in order to sell their flagship product, memory. To focus on microprocessors, he left to form Zilog in late 1974, where he went on to develop the highly successful Z80 that helped usher in the home computing revolution of the 1980s.
Author’s note: *An Intel C4004 processor with no gray marks, licensed under the Creative Commons Attribution-Share Alike 4.0 International License: https://bit.ly/3xHF58S
Stuart Cording is an engineer and journalist with over 25 years experience in the electronics industry. You can read many of his Elektor articles at https://www.elektormagazine.com/cording.
This article first appeared in the November 2021 special edition of Elektor.