The ACM’s Special Interest Group in Computer Architecture (SIGARCH) announced on June 20th, at the International Symposium on Computer Architecture (ISCA), that Professor Abhishek Bhattacharjee is the recipient of the 2023 Maurice Wilkes Award. The Maurice Wilkes Award is the most prestigious honor bestowed on a young computer scientist or engineer working in the field of computer architecture, where young is defined as being within 20 years of having joined the field. Bhattacharjee has received this honor for his “contributions to memory address translation used in widely available commercial microprocessors and operating systems.”
The award is named after Sir Maurice Wilkes, a pioneer of computing systems who made fundamental contributions to the underpinnings of computer architectures and computing in general. Although Yale has a history of breakthroughs in computer architecture via innovations in instruction-level processing and VLIW architectures, this is the first time an active Yale scientist has received this honor.
Bhattacharjee’s lab has focused for over a decade on the problem of memory address translation. Address translation has, for decades, simplified programming and permitted software developers to envision memory as a flat and linear array of bytes. This spares software developers from having to understand the complex reality of physical memory as an amalgam of various memory and storage devices. This distinction between “virtual” and “physical” memory is one of the central tenets of computer systems design. As described by Bhattacharjee, “…address translation is a step that all computer systems go through to convert the code that a programmer has written to the reality of moving data around in a computer. By having that mechanism in all systems, programming becomes way easier.”
As a consequence, all computer systems, from the smartphones in our pockets to the servers that form the backbone of Google’s, Meta’s, and Microsoft’s data centers rely on address translation. But, there is a tax in exchange for this simplification in programmability – a performance overhead to translate virtual to physical addresses. And, unfortunately, as modern software uses ever-increasing amounts of data, this performance tax has become pernicious.
Bhattacharjee’s specialty is reducing that tax as much as possible. And, his research has not only made its way from academia to industry, it has done so on a large scale. Bhattacharjee’s ideas can be found in AMD’s Zen central processing unit (CPU) cores, Nvidia’s graphics processing units, and the Linux operating system. By some estimates, his work has made its way into as many as 1 billion CPU cores and active Linux deployments. These systems, in turn, are at the forefront of powering the next generation of AI in our edge computing systems and data centers worldwide.
Bhattacharjee was drawn to the area of address translation because he found it challenging to understand in his days as a student.
“I had really struggled to make sense of this concept, and it’s so vital and fundamental to computing that it really bothered me that I didn’t understand it,” he said. “I started reading research papers to try to learn more about how this step in a computer works. The more I read, and the more that I discussed my findings with my Ph.D. advisor, Professor Margaret Martonosi at Princeton University, the more I realized that there seemed to be many open problems to future-proof memory address translation.”
Today, Bhattacharjee’s interest in memory translation extends beyond his research and goes into the classroom as well. Memory address translation features heavily in his courses on Introduction to Computer Systems and Computer Organization (CPSC 323) as well as Computer Architecture (CPSC 420).
“One of the things that I tell my students is that it’s completely OK if you’re struggling to understand this idea,” he said. “I was in their shoes, too; I didn’t really understand it, but I felt encouraged by many mentors, teachers, and collaborators to actually fix that lack of understanding, as opposed to feeling like there was some inadequacy in me. And I think that’s how this career arc of mine unfolded.”
Bhattacharjee described receiving the Maurice Wilkes Award as a “mind-bending honor,” noting that past recipients are among those who have shaped the foundations of computer architecture. Prior recipients have received this honor for contributions fundamental to computer systems, including (but not limited to) high-throughput out-of-order processors, high-performance caches, computer security, memory consistency models, on-chip networks, cache coherence protocols, multi-threading, power/temperature-aware computing, biologically-inspired computing, and much more. Several of the prior recipients have built some of the most notable computer architectures in history including Intel’s Pentium chips, AMD’s Athlon chips, DEC’s Alpha chips, and more.
“Technically, this award is given to an individual for a single impactful contribution to computer architecture,” he said. “But, in reality, awards like these happen only because of the many impactful contributions of many scientists in both academia and industry helping shape the recipient’s career.” For that reason, Bhattacharjee credits his mentors, collaborators, past and present for this honor.