Gerald Sussman

Ruse Laboratories
Mar 18, 2015 2:39AM


An eminent academic and influential author, Sussman is one of the world’s most prominent figures in the realm of computer science. A professor of electrical engineering  and computer science at MIT, Sussman is a co-author, with Hal Abelson, of the classic textbook Structure and Interpretation of Computer Programs and its attendant course of study. Sussman was an architect of Scheme, a principal programming dialect. He is a leading researcher in artificial intelligence, a field in which he has worked since 1964. Sussman has been a member of the board of directors of the Free Software Foundation since the organization’s inception.

About Scheme

In January 1978 many members of the MIT Electrical Engineering & Computer Sciences (EECS) department attended a class taught by Charles Botchek on the physical electronics of Very-large-scale integration (VLSI) technology.  I was one of those faculty members.  It was pretty exciting, because we felt that we could influence the way that discipline was evolving.  The class ended on 6 February 1978--the day of the famous Blizzard of '78-- so many of us were stuck at MIT for a while.

We had heard that Lynn Conway from Xerox and Carver Mead from Caltech were making real progress on making it possible for people who were not at a chip-fab facility to specify chips for experimental designs. In the Spring of '78 we invited Lynn Conway to teach the class on VLSI design that she was working on with Carver Mead.  Just a few years earlier Guy L. Steele Jr., (then my graduate student) and I invented a simplified but elegant version of the LISP family of languages that we called Scheme.  Guy and I wrote a number of internal memos (Lambda the Ultimate...) that later became famous.  Guy enrolled in Lynn's class. For his term project he designed and fabricated a direct interpreter for Scheme, called Scheme-78.  It didn't quite work (because of three missing wires); it didn't have a garbage collector; and it was too small to do anything impressive; but it encouraged us to try again. Over the next few months Guy Steele, Jack Holloway, and I designed a new interpreter that we thought could actually be run on a real memory and tested with real programs.  I designed the register array, Guy and I developed the microcode.  Jack made a PLA generator that could hold the microcode, and we roped Alan Bell of Xerox PARC into assembling the Scheme-79 Chip.  We pulled this off in a few man-months of time and it worked!  Scheme-79 had a mark-sweep garbage collector with a Deutsch-Schorr-Waite mark algorithm and a two-finger compacting sweep. It also had a two-level microcode: The main PLA contained rather high-level microcode instructions that were further elaborated by a nanocode PLA that operated the register array.

Further encouraged, I started a new project to make a chip that was actually big enough and fast enough to be useful to run real research programs.  This was the Scheme-81 chip.  It was a 32-bit machine, with 6 bits of type code and 26 bits of address.  It had microcode support for everything required to make a Scheme computer operating system, including a stop-and-copy garbage collector, a coprocessor bus, and an interrupt system. For Scheme-81 the microcode was written by Richard Stallman, Chris Hanson, and me.  (Steele had graduated and moved on to CMU as faculty.)

Howie Shrobe (my former student) and I led a pretty big effort to apply artificial intelligence technology to the computer-aided design of integrated circuits.  Students involved were Neil Mayle, John Batali, Ned Goodhue, and Jon Taft.  We developed LISPIC, a Lisp-based integrated-circuit design system that supported the design of Scheme-81.  I worked with Shrobe, Mayle, and Batali to build a Design-Procedure Language for describing the layout of chips in terms of high-level abstractions.  Batali wrote an elegant PLA generator; Shrobe wrote a data-path generator, and the Daedalus graphics-editing system for the MIT Lisp Machines, which we used for this project. Goodhue worked with me on the initial interpretation of the microcode, and he ported Clark Baker's design-rule checker to the LISPIC system. Mayle and Taft designed and built the boards that plugged into the Lisp Machine for testing the chips.  Batali did the final layout, routing, and assembly of the Scheme-81 chip using the LISPIC software.

The only run of Scheme-81 chips produced chips that worked, but unreliably, due to an over-long diffusion wire (that was my fault!). We got it to work by carefully adjusting the substrate bias, and it worked long enough to run the doubly-recursive fibonacci program at amazing speeds.  Phil Agre designed and built an ASSQ coprocessor chip to sit on the Scheme-81 bus and provide high-speed access to linear tables and lists.  It was apparent that with another run and a few tiny edits we could have high-performance Scheme machines.  But at this point we ran out of energy, so we declared a success and stopped building more of these chips.

Overall, this was a fun set of projects for the time (from 1978 to 1982).  However, it diverted me and many of my friends from doing some more fundamental AI work that we really wanted to do.

- Gerald Sussman

Gerald Sussman

Ruse Laboratories