原创 A look at the 4004, nanometre CMOS VLSI and MEMS logic (Part 1)

Jack Ganssle recently wrote a column that reminded me that this year is the 40th anniversary of the Intel 4004, the four-bit microprocessor that transformed the electronics industry.

His column also reminded me of that old saying: "what goes around, comes around"—issues that are facing embedded systems developers and electronics engineers today are more or less the same as the ones embroiling the electronics industry in the early 1970s when I first became aware of the industry's first fully integrated microcomputer on a single integrated circuit.

Then, as now, the industry was trying to find ways to take planar metal oxide semiconductor (MOS) technology used to build microprocessors and microcontrollers to the next step, either by finding ways to scale the underlying MOS logic structures to get more density, more performance, and lower power, or shifting to another approach that would allow them to do so.

In 1971, about the time the 4004 was introduced, I was a recent graduate of Columbia University and about a year into a job at the California Institute of Technology as associate managing editor of the in-house tech publication, Engineering and Science Magazine.

At that time I was working with Caltech Professor Carver Mead (co-author with Lynn Conway of the seminal book "Introduction to VLSI System Design"), helping him with an article for the magazine titled "Computers that put the power where it belongs."(pdf) The article was about how microprocessors like the 4004 were going to change computing and put computing power in the hands of ordinary people, either directly through what Dr. Mead called 'personal computing devices' or embedded in the existing machines, replacing the bulky inner electromechanical workings.

I still have a copy of one of the photographs (below) taken as a possible illustration for the article. In the photo, the 4004 chip that Dr. Mead passed over to me rests in my hand. As I held that tiny (by the standards of that time) chip, I was struck by the potential of the microprocessor revolution. I wanted to learn more, and signed up for every electronics class I could at the institute.

Bernard Cole gives the Intel 4004 a hand.

At the same time, I was helping some of the Caltech faculty and grad students in the beginning stages of proposal writing for funding from either the National Aeronautics and Space Administration (NASA) or the Defense Advanced Research Projects Agency (DARPA) on the three technologies that are still front and centre in the electronics industry: globally clocked synchronous logic, clockless asynchronous logic design, and microelectromechanical systems (MEMS). The first two originated in the EE/CS department, while the MEMS effort was the inspiration of some grad students in the physics department.

The aim of the first two groups was to find ways to move beyond the medium scale (MSI) and small scale (SSI) integration planar MOS circuits used at that time in most mainframes and minicomputers (as well as many missile and space craft electronics systems) to the next steps: Large Scale (LSI) and Very Large Scale (VLSI) Integration.

The two groups were split between the use of either synchronous MOS logic or asynchronous MOS logic. In simplistic terms, synchronous logic is globally clocked such that all transistors on a chip march in lock step and are always on and always consuming power. Asynchronous logic is clockless in the global sense. Instead, it is clocked locally such that transistor logic is turned on – and consuming power—only when needed. It is then turned off locally when it is not needed for switching functions.

Alternatively, the MEMS group wanted to take an end run around all of the scaling problems inherent in MOS circuits extended to very large scale integration (VLSI) densities. Inspired by the lectures by Caltech physicist Richard Feynman in the late 1960s, they wanted to build devices for storing information and performing logic functions by means of MEMS relays and switches that could be etched directly in silicon or any of a number of semiconductor substrates using the same fabrication techniques used to build MOS circuits.

The argument made by the micromachine proponents was that while MOS VLSI circuit techniques were the fastest way to get to the higher levels of integration that the DoD and NASA needed, both approaches were prone to a host of potential problems as transistor sizes were scaled to smaller dimensions. Synchronous logic would hit the wall first and while the use of asynchronous logic would extend the usefulness of MOS VLSI, it also would run into problems.

Unlike a purely mechanical/electromechanical system based on MEMS, their arguments went, software-programmable semiconductor devices are all prone to errors and to problems introduced by the environment: heat, temperature, vibration, shock, dust, humidity, x-rays, and radioactive particles. And as size of transistors are scaled down all sorts of secondary and tertiary problems relating to reliability, power, voltage and noise issues get worse.

I left Caltech shortly thereafter to take a job as Silicon Valley Bureau Chief for the now defunct Electronics Magazine. So I lost track of how successful the proposals submitted were or even if they even got to the final submission stage, but in retrospect it is obvious that for the electronics industry in general, the fast track argument won the day.

[To be continued at A look at the 4004, nanometre CMOS VLSI and MEMS logic (Part 2)]