Prof. Trevor Mudge retires after 46 years in computer architecture
Trevor N. Mudge, the Bredt Family Professor of Engineering, retired from the EECS faculty on May 31, 2023, after 46 years at the University of Michigan. His career has been marked by significant contributions that have helped to shape the evolution of computing. His contributions included research into architectures for high performance computers and processors, the potential offered by parallel computing architectures, and programming languages. Most notably he contributed numerous research projects that led to techniques for energy reduction in computer systems ranging from large data centers to small portable system-on-chip designs. His work has greatly influenced both research literature and commercial products.
Trevor grew up in the East End of London, an area that can be politely described as economically depressed. He graduated from high school early with mathematics and physics concentrations. At that time and place the smartest students studied classical Greek and Latin. Mathematics and physics were considered a poor second. Additionally, the people in his neighborhood were nonplussed by the idea that anyone would want to go to school past age 16. As a result Trev didn’t picture himself going to university. However, he had learned of a program in Cybernetics in the physics department at Reading University and successfully applied. He earned his Bachelor of Science with Honors in Cybernetics, and a Mathematics Minor.
After graduating from Reading he decided to go abroad for graduate school. Because he spoke only English, he applied to graduate school in the US. After working at a variety of jobs – delivering cars across the US was the most memorable – he enrolled at the University of Illinois in Urbana-Champaign where he pursued a PhD in Computer Science. He worked on a wide range of topics including computer architecture, image processing, parallel processing, and fault tolerant computing. A distinguishing feature of Illinois was that prototypes were built there, thus from early in his career he was engaged in fabricating computers. Several of his early projects involved ILLIAC III – an early image processing computer. Finally, he settled on the idea of digital systems design languages (today’s reader will think of languages like Verilog). His PhD dissertation thesis was “A Computer Hardware Design Language for Multiprocessor Systems.”
In 1977, after completing graduate school, Trevor joined the faculty at the University of Michigan in the Electrical Engineering department, where his work initially focused on high performance computers, and the potential offered by parallel computing architectures.
The Electrical Engineering department became known as the Electrical and Computer Engineering Division in the EECS Department, but as Trev recalls, “Dan Atkins was the only other computer faculty member in the department at that time. Fortunately, he became an early mentor and was immensely helpful. A few other helpful faculty members, who were re-educating themselves to be computer types also offered a lot of help, especially Dick Volz.” At that time Trev developed several of the early computer architecture classes, for example ECE 365, Introduction to Computer Architecture (now known as EECS 370). He also developed the VLSI design classes in conjunction with faculty members Ron Lomax, Keki Irani, and Dan Atkins. This was inspired by Lynn Conway’s pioneering work to make VLSI widely available to university students and faculty. As a result of this work he won the college’s outstanding teaching award in 1983.
In 1986, Trevor, along with his U-M collaborators John Hayes and Quentin Stout, and two collaborators from nCUBE Corporation, spent a considerable amount of time working on and trying to program an early, 64-core parallel computer. It was called the nCUBE and was hypercube connected. It was the basis for numerous papers on parallelism and collaborators summarized their experience in an IEEE MICRO paper entitled “A microprocessor-based hypercube supercomputer,” which won the Best Article of the Year award.
As a result of the start of VLSI activity, Trevor began investigating the design and fabrication of integrated circuits, particularly very fast CPUs fabricated from Gallium Arsenide, which supported clock rates of three to four times faster than silicon. According to Trev, “Despite their speed advantages, we also found Gallium Arsenide CPUs to have limitations, but some outcomes of this work were new algorithms for placing latches in digital systems to maximize frequency, and the need for the much deeper pipelines now common in today’s high-end processors.”
In 1997, he co-authored the first paper to consider pre-execution of instructions, which improved control flow predictions and data cache performance. That same year, he received the College of Engineering’s Research Excellence Award.
In the late 1990s, Trev shifted his focus from the development of high-performance architectures and processors to the pursuit of new technologies for ultra-low-power computing. This was partly because he realized that low-energy computing would enable a new class of low power devices. But there was also the appeal of tackling a new and interesting problem – the building of low-energy architectures.
Trev’s interest in low-power computing was the result of observing that the quest for higher clock rates was unsustainable due to excessive power consumption. He advocated for power consumption as a design limit for computer architectures in the article “Power: A first class design constraint,” which was published in the April 2001 edition of Computer, and which has been widely cited. The idea, which was that power is a design constraint not only for portable computers and mobile communication devices but also for high-end systems, and that the design process should not subordinate power to performance, seems obvious now but at the time it was a novel position.
In recognition of his accomplishments, Trev was recognized with the Bredt Family Endowed Professorship in Engineering at the University of Michigan in 2002.
In 2006, at the invitation of semiconductor company ARM, Trev, along with fellow EECS faculty members Todd Austin, David Blaauw, Scott Mahlke, and former faculty member Thomas Wenisch, began a U-M/ARM research partnership. Trev served as director of the U-M center. That ran for a remarkable 15 years, through 2020, and the partnership produced a trove of intellectual property, much of which was targeted at saving energy in today’s smartphones and other personal, portable electronics.
According to Trev, one of his most important accomplishments has been to mentor and collaborate with graduate students. All in all, he has supervised 60 doctorates and many masters degrees, as well as having provided mentorship to numerous undergraduates. Many have gone on to make significant impacts of their own on the field. When asked what advice he has for student researchers, Trev says, “Go where you want to go and do what you like to do. When thinking through an idea suspend disbelief.”
“Trev was an absolute joy to have as an advisor,” says Jonathan Beaumont, Trev’s former student and now a faculty member at CSE. “He pushed everyone in the lab while never losing his dry, sardonic wit. Last minute pushes for paper deadlines, rather than feeling arduous, were fun social gatherings to look forward to. He helped me find success in my own way, and I don’t think I’d be where I am today without his mentorship.”
Rich Uhlig, now an Intel Senior Fellow, Corporate VP, and Director of Intel Labs, was another of Trev’s students. He says, ‘Trev has an amazing command of the history of computing systems. He taught me the importance of understanding what has succeeded and failed in the past, and most importantly… why? That can be the key to making the next big contribution to the field. It’s a lesson that I’ve carried with me my entire career.”
Jeff Ringenberg, now a faculty member at CSE, shares his thoughts: “Trev provided me with an incredible amount of academic, moral, and financial support throughout my many years of grad school. His quiet guidance, sage advice, and British humor made the time an immensely enjoyable experience and gave me the freedom to ultimately discover my passion for teaching which continues to enrich my life to this day.”
A few notable projects that Trev contributed to included Vertigo, which was adopted by ARM as their Intelligent Energy Management system; Razor, which provided automatic timing violation detection to support lower power operation; Drowsy Caches, which allowed caches (the largest part of most chips) to run at significantly lower power levels; runahead computing that pioneered the idea of pre-execution to improve control flow and memory performance; work to create 3D processor architectures, in which a number of processor cores are interconnected on and between a number of layers of silicon—this approach is now appearing as “chiplets” that can be stacked in 3D; and work that enabled a programmable architecture to support a range of wireless standards in a single mobile device. Many of these concepts were verified by experimental prototypes.
Regarding Trev’s research output, his former student Don Winsor, now head of the EECS Departmental Computing Organization and an adjunct faculty in the department, says, “Trev always amazed me with his persistence and his resilience. If things weren’t going well with a research problem, he always encouraged his students to try other approaches; to look at things in a different way. He could break the ice in difficult situations with his ironic sense of humor.”
In 2014, he was awarded the ACM/IEEE Eckert-Mauchly Award, which is widely viewed as the computer architecture community’s most prestigious recognition, “for pioneering contributions to low-power computer architecture and its interaction with technology.” Also in 2014, he received a Distinguished Achievement Award from the University of Illinois Computer Science Department for his work to advance the field of low-power computer architecture and its interaction with technology.
Over the years, Trev has published over 400 technical papers, a number of which have been recognized with best paper and top pick awards. Four of them have been selected for high impact, “test of time” awards. He holds 45 patents and is a Fellow of the Institute of Electrical and Electronics Engineers (IEEE) and the Association for Computing Machinery (ACM).