In memory computing panel during the VLSI-SOC conference

Wed, Oct 10, 2018

I was invited by Said Hamdioui to participate to a panel session during the 2018 VLSI-SOC conference in Verona, Italy. The participant to the panel were :

• Ian O’Connor (Univ. Lyon, France)
• Mehdi Tahoori (Karlsruhe Institute of Technology, Germany)
• Henri-Pierre Charles (CEA, France)
• Lionel Torres (Lirmm, France)

Thanks to the great organization and the different panelist scientific culture, it was a great opportunity to discuss our different view and with the audience.

The question list proposed by Said were :

1. Unifying terminology: What does computation-in-memory mean? What is the main difference between other computation paradigms (near-memory  computing, conventional computing, etc) and other terminologies (processing in memory, logic in memory, memory computing, ..)
2. Myth or reality: What makes computation-in-memory feasible or not feasible? It is really coming? Or just a hype?Potential of Computation-in-Memory: Assume it is a reality, what kind of applications can/ cannot benefit from such new  architectures? How big could the potential be (e.g. in energy and energy efficiency, throughput, Performance energy efficiency, …)
3. Technology aspects: What kind of emerging memories could make computation-in-memory feasible? Why? Why not? What about the control circuitry?
4. Paradigm change: How will computation-in-memory change (not change) the way we used to design our circuit and computers? Programming models? etc
5. Challenges: What are the major challenges that should be solved before having computation-in-memory become a reality? What is your expectation? When these computer will become a reality?