Rump Sessions (Tuesday, June 27)
8:15 PM (Location: McKenna Hall Auditorium)
Next 10 years for device research: Transistors and beyond
Mathieu Luisier (ETH-Zurich)
Keisuke Shinohara (Teledyne)
Suman Datta (U. Notre Dame)
Mark Rodwell (UC-Santa Barbara)
Arijit Raychowdhury (GA Tech)
Steve Koester (U. Minnesota)
Hidenobu Fukutome (Samsung)
DRC has witnessed key innovations in transistor research during its 75-year history. During this time, the continuous increase in CMOS density has enabled the emergence of today's digital society. However, further breakthroughs may be hindered by the slowing and impending end of Moore's scaling law. How much life is left in Si CMOS scaling and what device(s) will drive the next electronic revolution? In this rump session, possible evolutions of transistor research for the next 10 years will be discussed by a panel of experts from industry and academia. The focus will not only be set on the future of Si logic switches for digital applications, but also on alternative materials, structures, and areas where significant progress can still be made, such as terahertz electronics, in an attempt to elucidate the path for future device research.
• What will the key materials and device architectures be for transistors over the next 10 years?
• Will transistor scaling cease to be effective due to interconnect parasitics, manufacturing costs, etc.? How much life is left in Si (both in terms of scaling and usage)?
• Is replacing Si in MOSFETs (or FinFETs, etc.) worth the cost or does the next revolution need to be with entirely new material and device operation (e.g., TFET, NC-FET)?
• Can the decline in transistor-level performance improvement continue to be compensated at the circuit level? For how long?
• What’s beyond transistors? What should new graduate students work on?
8:15 PM (Location: McKenna Hall 210-214)
Neuromorphic computing -- Do devices matter?
Rashmi Jha (Univ. Cincinnati)
Bipin Rajendran (NJIT)
Wilfried Haensch (IBM)
Kaushik Roy (Purdue)
Yiran Chen (Duke)
Rajit Manohar (Yale)
Machine learning algorithms inspired by the key organizational features of the neural networks of the brain have revolutionized how we interact with artificial computing systems today. However, their widespread adoption on mobile and IoT platforms faces challenges because of the large network sizes and extremely long training times involved in learning underlying features of user-dependent data. Further, there is also widespread research to build cognitive computing systems that more closely mimic the time-based information encoding and processing aspects of biological neurons and synapses.
Hence, special purpose accelerators, either based on conventional CMOS or emerging post-CMOS devices, have been proposed for efficiently implementing these algorithms, as well as for realizing bio-mimetic cognitive computing systems. In this rump session, a panel of experts from industry and academia will discuss what should materials/device engineers focus on in the quest to build cognitive systems with human level intelligence and efficiency.
• What are the major challenges for neuromorphic computing?
• What is the role of devices in enabling neuromorphic computing?
• What should the device engineer/researcher be working on for the neuromorphic computing space?
• What are the key features/metrics for an ideal device?
• Do you think emerging devices based, hardware-enabled neuromorphic computing will be able to compete with CMOS-hardware-enabled neuromorphic computing?
• How do you see conventional digital computing interfacing with neuromorphic computing?
• What is more important--devices for second generation networks or devices for spiking networks?