Biruk Mammo Awarded Rackham Predoctoral Fellowship for research Into robust processors

Biruk Mammo Enlarge
Biruk Mammo

Biruk Mammo, a graduate student in the Computer Science and Engineering program, has received a Rackham Predoctoral Fellowship to support his research while he completes his dissertation entitled, The Brain-like Microprocessor: Self-organizing, Redundant and Reliable. The Rackham Predoctoral Fellowship is awarded to outstanding doctoral candidates in the final stages of their program who are unusually creative, ambitious and risk-taking.

Biruk’s research focuses on solutions to assist processor designers in addressing the challenges in reliability that arise as processors become ever more smaller and more densely packed with transistors.

According to Biruk, “Innovations in microprocessor design have placed computers in the palms of our hands, with several orders of magnitude more performance than the room-size computers of the early days. Unfortunately, with that decreasing size, the transistors that are the fundamental building blocks of microprocessors are becoming more and more fragile. For this reason, many in the industry expect that in ten years’ time we will have microprocessors where at least one transistor breaks every day. If this happened today, customers would have to replace their microprocessors daily!” As a result, computer architects must radically alter the way they think about designing microprocessors using unreliable fabrics.

Biruk’s dissertation explores a novel design approach in that direction. This approach leverages the abundance of transistors in a microprocessor to create redundant components capable of re-organizing themselves around transistor failures as they occur, to realize microprocessors that operate correctly, even in the presence of many transistor failures. The capability to self-organize – one of the marvels of the brain – also brings an added benefit: these brain-like microprocessors can adapt their internal configurations to provide the best possible performance for whichever task their users want them to complete.

Biruk Mammo is advised by Prof. Valeria Bertacco.