First-Hand:Invention of an Integrated Circuit
Invention of an Integrated Circuit
Should my grandson ask what I did when I first got out of college, I would say that "50 years ago this summer I invented the integrated circuit!" "Wow, Grampy, that's cool. But why aren't you famous?" "It's because I used the wrong technology - superconductivity instead of semiconductivity." As we approach the fifty year mark of semiconductor integrated circuits, it may be of general interest to hear one engineer's story from that time and milieu of another kind of integrated circuit.
It begins with me a year out of MIT in the summer of 1956; I was working with analog computers at GE's General Engineering Lab in downtown Schenectady and looking for something new to do. A senior engineer showed me Dudley Buck's cryotron article; Buck had systematically cataloged various physical transitions in electrical conductivity. He noted the change of a superconductor to a normal conductor with the application of a magnetic field, and invented a switching device called the cryotron to utilize this effect. It consisted of a gate wire with a low critical field, which could be switched from the superconducting state to a resistive one by the application of current to an insulated control coil wrapped around the gate. He used tantalum and niobium for the wires. The device could have a current gain, and as it had two terminal pairs like a relay would be easy to make logic and memory circuits of any complexity.
This greatly interested me; it seemed something that could get me into digital computers. I knew that GE's Research Lab (then a separate organization) out in the Schenectady countryside could liquefy helium, needed for the very low temperatures at which cryotrons operated. Some managers at the Research Lab had heard a talk by Buck and were interested in any way they could help GE's new Computer Department in Phoenix, and were therefore receptive to my volunteering to try some experiments using their facilities.
So started three years of great fun, sometimes all by myself and sometimes with help, sometimes as a bootleg project and occasionally with some funding. In October 1956, I first duplicated Buck's wirewound devices: flip-flops and memories. During 1957, I changed the materials to the much more tractable tin and lead. The next improvement was to make the gate a cylindrical thin film to get more resistance and therefore more speed. As people joined and left the project, experimental progress was slow, plagued with contact resistance and insulation breakdown.
What was clear to me was that digital computers would need great quantities of switching and memory devices. Using germanium transistors seemed difficult. They had to be made and tested one at a time, with great numbers of rejects. They then had to have their little wires stuck into a circuit board and soldered up. The circuit boards had to connect with other circuit boards and connect to memories also made of little individual devices. To make progress something had to be done about all these interconnections. Wirewound cryotrons though would be just as hard to hook up into large arrays as transistors.
I think many saw that some devices would have to be invented so that the interconnections could be made at the same time as the devices, and that these would have to be essentially two-dimensional. The back rooms of AIEE-IRE conferences were full of ideas. One I remember was the Thin Film Transistor at RCA Labs. I think it failed because the yield of devices with good switching characteristics was much too poor. Tunnel diodes, magnetic amplifiers, pneumatics, and other semiconductors beside germanium were considered. So cryotrons didn't seem so bizarre, despite the need for really low temperatures. Even John D. MacDonald, in a cold war novel had a Travis McGee-like character say, "When you change the conductivity of one of the ferride plastics, what effect would that have on the reliability of adjacent transistors, diodes, cryotrons, masers, parametric amplifiers and so on?"
Cryotrons offered significant possibilities. They were made of amorphous metals, easily deposited (no need for crystalline structures or diffusions). The same metals could be used for interconnections. No other devices (capacitors or resistors) were needed. Switching characteristics seemed consistent, the devices had gain, they had independent terminal pairs (like relays). Could they be made essentially two-dimensional, so both the switching and memory devices could be made together with all the interconnections?
I knew that the gate could be a flat strip just as well as a cylinder, and it eventually became clear to me that the control "coil" could also be flat, lying on top of the gate. On July 27 I went over to GE's wholly-owned printing company on the other side of town to order some flat controls made with ordinary photolithography.
So there was my invention in the summer of 1957: a computer device suitable for both logic and memory, where the interconnections between devices could be made with the same materials and the same fabricating processes as the devices themselves, on some sort of flat substrate. In fact, an integrated circuit! I published some papers about this. I got the controls in September, but it wasn't until December of the following year that devices with a gain greater than one were made. Of course, it is possible somebody else was there before my invention (or extension of Buck's invention if you will). Researchers at IBM and Arthur D. Little were working on superconducting devices for logic and memory including cryotrons, and so maybe were others.
My enthusiasm over the possibilities must have been contagious because in November of 1957 a new hire to the Research Lab, Vernon L. Newhouse, was convinced to join the project. We had a close collaboration for the next two years, writing many papers and patents, alternating as first authors. Much progress was made in understanding the physics involved which allowed engineering calculations to be made on gain and speed, but experiments were plagued with poor film quality and insulation shorts. Not until June of 1959 did we manage to get a shift register working. I wrote a fairly elementary book for McGraw-Hill in 1959. Newhouse later wrote a much more complete one.
In May of 1959 engineers from the Computer Department started getting involved. A computer lab was being set up in rented quarters in Palo Alto, across the street from Schockley Labs (it later moved to Mountain View and then to bespoke facilities in Sunnyvale). Most of the staff were veterans of the first transistor banking computer, built by GE for Bank of America. I accepted their offer and moved there in January 1960; Newhouse with others continued working in Schenectady.
Thus started another five years of fun with the great team that was gradually formed in the Bay Area (not yet called Silicon Valley). The first goal was to make a small demonstration computer to show practicability. Known auxiliary problems such as input/output devices and low-temperature refrigerators would be dealt with later. Speed was not emphasized as it would naturally increase as fabrication dimensions were decreased. Progress was slow however. It took nearly all of 1960 to improve on the work in Schenectady. Gates were deposited through masks, leaving photolithography of that layer to do later. In retrospect, the fabrication problems were probably all due to contamination - oxidation. poor adhesion, faulty insulation, spotty photolithography. Nevertheless, in 1961 demonstration memory and shift register circuits were made to work.
A complex array called the Arithmetic Unit was designed and in November the first unit was made, although the first perfect one wasn't until February of 1962. This single array had 6 bits of arithmetic and registers, input, output and some control circuits - about 330 logic gates altogether. Is this the first "integrated circuit" to be made of that complexity? I don't call it a "chip", because the substrate was a big glass plate. Minimum dimensions were thousandths of an inch, not thousandths of a millimeter.
The demonstration computer was put together towards the end of 1962, but very little worked. It had five arrays, the AU plus memories and more control circuits. Another one in April of 1963 worked better (I have it in my study), and a third in July was perfect. Was this the first integrated circuit computer?
An intriguing feature of cryotrons is that if you push current into one of two (or more) parallel paths, it will stay there - superconductivity really does mean zero resistance. Thus memory is inherent in any circuit; all of us in the field felt we should take advantage of this. In Sunnyvale we designed content-addressed memories and got some working in 1963. A 17 word by 36 bit array had some 2000 devices! A record?
As a side story for those who think personal computing is new, we had personal terminals in our offices (and placed in a few homes). In 1962 I could send and receive mail messages with colleagues in Phoenix, launch scientific calculations and get the results almost instantly, do simple word processing, and wrote a little AI program for fine tuning expense reports.
What else was happening in computer devices in the early 60's? We knew from conferences about Jack Kilby's work at TI, but didn't think it attractive. It was a bar of semiconductor with multiple diffusions to get the different devices and needed little wires to connect the devices - still a three-dimensional array. We knew about the work of Bob Noyce at Fairchild and of others in the Bay Area; there were local AIEE-IRE meetings and we all went to the same restaurants and bars. Jay Last's group at Fairchild was working on the same messy fabrication details as we were, but we thought we had a better device.
However, sometime in 1964 I saw the importance of Jean Hoerni's planar silicon transistor. It also consisted of a few basically two-dimensional layers and could be made at the same time as its interconnections. And unlike cryotrons it worked at room temperature. Realizing that Fairchild or others would eventually get it working, I could see the end of cryotrons. I left Sunnyvale in 1965 to pursue other interests, as the phrase goes - logic and systems design for GE in Phoenix. Yet as I said in a retrospective some years later, if cryotrons worked at room temperature they'd be found all over by now!
Although "my" technology didn't win and I'm not famous, it's gratifying to realize that the basic scheme I worked out with my colleagues is still used, fifty years later, by the semiconductor integrated circuit industry. The numbers are hugely different, but it is still the vapor and chemical deposition of a few layers on a flat substrate, each patterned by photolithography to make an array of functioning devices at the same time as their interconnections. Connections off the array are made at the edges. Protective layers keep out contaminants. I'm not implying that others borrowed our work but rather that the design choices we made were also obvious to them.
It's necessary to make clear again that this is a personal history. I have not mentioned the names of my many co-workers, to whom I apologize and owe thanks. It would be interesting to me to hear the stories of others working on superconducting or semiconducting devices during the late 50's and early 60's, even at the expense of finding out that my invention was anticipated or my recollections wrong.
- ↑ Buck, D.A.: Proc. IRE, vol. 44. p.482, 1956
- ↑ MacDonald, John D., Area of Suspicion, Fawcett, 1954, revised 1961, p. 192
- ↑ Bremer, J. W.: Joint AIEE-IRE-NYU Conference on Cryogenic Electronics, New York, Dec. 9, 1957
- ↑ Bremer, J.W., Elec. Mfg., vol. 61, p. 78, February 1958
- ↑ Newhouse, V. L. and J. W. Bremer, J. Appl. Phys. vol. 30, p. 1458, Sept. 1959
- ↑ Bremer, J. W., Superconductive Devices, McGraw-Hill, 1962
- ↑ Newhouse, V. L., Applied Superconductivity, J. Wiley, 1964
- ↑ Fruin, R. E., Oka, A. K., and Bremer, J. W., A Hybrid Cryotron Technology, IEEE Transactions on Magnetics, Sept. 1966, p.381
- ↑ Moss, M. J., letter, IEEE Life Member Newsletter, April 2007
- ↑ Lecuyer, C., Making Silicon Valley, The MIT Press, 2005
- ↑ Bremer, J. W., What Went Wrong With Cryotrons, IEEE Computer Group News, March 1969, p. 31