Lynn Conway and the VLSI Revolution

Abstract

Lynn Conway is one of the most consequential engineers in computing history and one of its most invisible, a combination that was not accidental. In 1968, she was fired from IBM when she announced she was a transgender woman and would be transitioning. Her pioneering work on dynamic instruction scheduling, done at IBM, was suppressed. She rebuilt her career under a new identity, and by 1980 had co-authored the textbook that created modern chip design (“Introduction to VLSI Systems” with Carver Mead) launching a revolution in semiconductor engineering that made the personal computer possible. Her story was unknown for thirty years; the acknowledgment of her contributions is still incomplete.

Lynn Conway, 2006. Image: Charles Rogers, CC BY-SA 2.5, via Wikimedia Commons.

IBM and the Suppressed Invention

Lynn Conway was born on January 2, 1938, in Mount Vernon, New York, and died on June 9, 2024, in Jackson, Michigan, at the age of 86. She enrolled at MIT in 1955 but withdrew in 1959 in psychological distress, eventually completing a degree at Columbia University’s School of Engineering and Applied Science in 1962 and a master’s degree at Columbia in 1963.

She was hired by IBM Research in 1964. At IBM, Conway made a fundamental contribution to computer architecture: dynamic instruction scheduling, a technique that allows a processor to execute instructions out of their original program order, identifying instructions that can be issued simultaneously and doing so, dramatically improving throughput. The technique is foundational in modern processor design, every high-performance processor built since the late 1980s uses out-of-order execution based on ideas Conway developed.

In 1968, Conway told IBM she was transsexual and would be transitioning. IBM fired her. The work she had done on dynamic instruction scheduling was suppressed, not published, not credited to her in IBM’s subsequent processor designs that used the technique, not acknowledged for decades. Separately, IBM’s Robert Tomasulo published the “Tomasulo algorithm” for dynamic scheduling in 1967, which received substantial recognition, while Conway’s own work on multiple-issue dynamic instruction scheduling, done on IBM’s Advanced Computing Systems (ACS) project, went uncredited for decades.

Conway transitioned in 1968, legally changed her name, and began her professional life again from scratch under a new identity. Her IBM career, her publications, her professional record, everything she had done before 1968 was gone. She could not use it without revealing that she was transgender, a disclosure that would have ended her new career as surely as it had ended the old one.

Xerox PARC and the VLSI Revolution

Conway joined Xerox PARC in 1973 as a research staff member. PARC was the most productive research laboratory in computing, the Alto, Ethernet, laser printing, and the GUI all emerged from it in the same period. Conway worked on computer architecture and, increasingly, on the question of how integrated circuit design could be made accessible to people who were not specialists in semiconductor physics.

The problem: designing VLSI (Very Large Scale Integration) chips (chips with tens of thousands of transistors) required deep expertise in semiconductor fabrication physics that only electrical engineers with specialized training possessed. The design process was a bottleneck: ideas for what chips should do existed widely among computer scientists, but the ability to turn those ideas into actual silicon was held by a tiny priesthood. The result was slow, expensive chip design that could not keep pace with the ideas that needed hardware.

Conway’s insight, developed with Caltech professor Carver Mead, was that chip design could be separated from chip fabrication by establishing simplified design rules that guaranteed correct electrical behavior without requiring designers to understand the physics of transistors. If a designer followed the rules (specified minimum widths, spacings, and layer relationships) the resulting design would work. The rules abstracted the physics away.

Mead and Conway developed these simplified design rules through 1977 and 1978, tested them with students at MIT and Caltech, and published the results in “Introduction to VLSI Systems” (1980, Addison-Wesley). The textbook became the foundation of an entire educational transformation: for the first time, computer scientists who were not electrical engineers could design custom chips. Universities could teach chip design as a software engineering discipline, not a physics specialty.

The ARPA VLSI Program and MOSIS

The textbook alone was not sufficient. Designing a chip was useless if there was no way to fabricate it affordably. Conway recognized that chip fabrication required access to a semiconductor foundry, a resource that small research groups and universities could not afford individually.

She architected a solution: ARPA (the Advanced Research Projects Agency, later DARPA) funded a program in which university chip designs were aggregated and sent to commercial foundries for fabrication together, sharing the costs. This multi-project wafer approach, later institutionalized as MOSIS (Metal Oxide Semiconductor Implementation Service), allowed university research groups to get chips fabricated at a fraction of the cost that individual runs would require.

The first ARPA VLSI Program chip run in 1979 produced chips designed by students at universities across the country, fabricated at commercial foundries using the Mead-Conway simplified design rules. The chips worked. The demonstration that universities could design chips, send designs to a shared fabrication service, and receive working silicon transformed chip design from a closed industry activity into an open educational and research discipline.

The consequences were structural. Fabless semiconductor companies (companies that designed chips without owning fabrication facilities) became possible: they could design chips using the established tools and methodologies descended from Mead-Conway, then send designs to a foundry for fabrication. Companies like Qualcomm, ARM, NVIDIA (in its early years), and dozens of others were founded on this model. The separation of chip design from chip fabrication that Conway architected is the organizational basis of the modern semiconductor industry.

Thirty Years of Invisibility

Lynn Conway did not reveal her transgender identity professionally until around 1999–2000, three decades after her transition. During those years, she built a distinguished career at Xerox PARC, then as a professor at the University of Michigan (1985–1998), working on computer architecture, VLSI, and later robotics. Her work was recognized and respected; her pre-1968 career was unknown.

Around 1999, her suppressed IBM work began to resurface as historians documented the long-classified Advanced Computing Systems (ACS) project: Mark Smotherman’s ACS history research and a July 1999 ACS reunion that Conway attended threatened to connect her pre-transition and post-transition identities, which she had kept separate for thirty years. Rather than let the link be exposed on someone else’s terms, Conway chose to disclose her history publicly herself, around 2000.

The disclosure revealed the suppressed IBM work on dynamic instruction scheduling, the erasure of her contribution to the Tomasulo algorithm’s underlying concepts, and the full arc of a career built twice, once destroyed by institutional discrimination, once rebuilt with the discrimination hidden.

Recognition followed, incompletely. The Computer History Museum honored her in 2014 with a fellow distinction. The IEEE Computer Society gave her its Computer Pioneer Award in 2009. The University of Michigan awarded her an honorary doctorate in 2018. ISCA (the International Symposium on Computer Architecture) retrospectively recognized her contribution to dynamic instruction scheduling.

In 2020 (fifty-two years after firing her) IBM formally apologized. At a company event, senior vice president of human resources Diane Gherson and CEO Arvind Krishna apologized to Conway on behalf of the company, and IBM presented her with its Lifetime Achievement Award for her body of technical work, including the dynamic instruction scheduling she had developed on the ACS project before her dismissal. The apology was a striking institutional reversal: the same company that had erased her had now, half a century later, named what it had done and honored what it had taken from her.

Info

The dynamic instruction scheduling work remains incompletely credited in the technical record. IBM itself reversed course in 2020, apologizing and granting Conway a Lifetime Achievement Award that explicitly named her ACS work, but in the classroom the gap persists: the Tomasulo algorithm is taught in computer architecture courses without standard reference to Conway’s parallel development. Correcting historical attribution is slower and harder than original attribution, and the work of revision is ongoing.

Legacy

Lynn Conway’s career is an argument against single-variable analysis of technological history. The story of who built what technology cannot be separated from the story of who was permitted to build it, whose work was credited, and whose identity made them invisible even when their contributions were foundational.

The practical legacy is clear: modern processors use dynamic instruction scheduling; the fabless semiconductor industry uses chip design methodology descended from Mead-Conway; every student who learned chip design from “Introduction to VLSI Systems” or its successors learned from Conway’s work. The companies that built the devices in everyone’s pockets were founded in a world Conway’s VLSI revolution made possible.

Conway died on June 9, 2024, at her home in Jackson, Michigan, at the age of 86. By then she had spent her final decades as an open advocate for transgender rights as well as an engineer, having lived, in the end, the full recognition of both halves of the career that institutional discrimination had once forced her to keep apart.