Just recently, I was invited to present a paper at a joint meeting of the International Technology Roadmap for Semiconductors (ITRS) and The Institute of Electrical and Electronics Engineers (IEEE) Rebooting Computing group. Each plays a critical role in bringing together academics, industry researchers, and government agencies to help shape the development of, respectively, semiconductors and computer architecture.
We met over a weekend at Stanford University in the hope of pioneering new approaches to computing in the face of the impending end of semiconductor scaling and guided by the notion that we can't rely on our past trajectory to plot our future course. To paraphrase Dr. Carlo Gargini, founder and past chair of the IEEE Conference on Intelligent Transport Systems, "We used up twenty years worth of scaling techniques in ten years chasing after server performance, and we knew we were doing it at the time, and so now we have to ask ourselves what do we do next?"
The presentations offered some fascinating perspectives on how things might change. Key topics and trends discussed included innovations in machine and natural learning, data intensive computing, trust and security, and memory–driven computing. But we were all promoted by the same question: what will happen when transistors stop getting smaller and faster?
My talk, titled “Memory–Driven Computing,” explored a phenomenon our research and engineering teams at HP have been observing: that conventional compute is not getting faster, essentially because the technologies we've been optimizing for the past 60 years to create general purpose processors (copper connectors, tiers of memory and storage, and relational databases) are all at their limits. It's not just a matter of the physics of CMOS devices, we need far more fundamental change than that.
Additionally, I noted, just as technology improvement curves are flattening out, data volumes are set to explode. Human-generated data will soon be too big to move even with photonics, and will thus be all but impossible to securely analyze in real time. Fundamentally, we’re reaching an inflection point. It may not happen tomorrow, but it’s only a matter of time before we do.
In response to this challenge, HP has been collaborating with IEEE Rebooting Compute and the ITRS, looking at the problem from a holistic perspective and taking into account both the evolutionary and the revolutionary approaches we need to accelerate scientific discovery and insights in the field. Our vision is that the different constituencies that we represent can change physics to change the economics of information technology, catalyzing innovation in successively larger circles.
Just bringing together the leading minds of semiconductor physics and computer architecture isn't sufficient, however. We need to bring an even broader perspective to the table, with more engineering and scientific disciplines represented because there are no more simple answers.
As I shared in my talk, HP Labs’s major research initiative, The Machine, has been examining these questions for several years now with the ambitious goal of reinventing the fundamental architecture of computers to enable a quantum leap in performance and efficiency, while also lowering costs over the long term and improving security. Almost every team within HP Labs is contributing to this effort, along with engineering teams from across the company’s business units. Central to our work is a shift from computation to memory as the heart of information technology.
The Machine will fuse memory and storage, flatten complex data hierarchies, bring processing closer to the data, embed security control points throughout the hardware and software stacks, and enable management and assurance of the system at scale. It may seem counter-intuitive, but by concentrating on massive pools of non-volatile memory, we expect to spur innovation in computation by allowing many different models of computation to work on the same massive data sets. Quantum, deep neural net, carbon-nanotube, non-linear analog – all of these models could be working in concert, connected to petabytes and exabytes of information derived from a world of intelligent devices.
By collaborating with the proven leadership of the ITRS and IEEE, we believe we can broaden, as we must, the impact of the technical innovations that we’re developing with The Machine. The result, we hope, will be the development of new ways to extract knowledge and insights from large, complex collections of digital data with unprecedented scale and speed, allowing us to collectively help solve some of the world’s most pressing technical, economic, and social challenges.