E2CDA: TYPE 1: Durable, Energy-Efficient, Pausable Processing in Polymorphic Memories (DEEP3M)
Cornell University, Ithaca NY
Investigators
Abstract
In modern computer systems, memories located close to the processing unit must be fast with nearly infinite endurance to support operation rates exceeding a billion per second. However, these memories cannot be scaled to very small sizes, i.e. they are low-capacity, and/or they lose their contents when the power is off, i.e. they are volatile. Existing high-capacity non-volatile memories, such as solid-state hard drives, must typically be situated away from the processing unit. As a result, it takes extra time for a processing unit to fetch data, process them and store them back. Furthermore, most non-volatile memories can only be erased and written a finite number of times. An ultimate memory should be suitable to be embedded in all systems. The desired features of this memory include non-volatility, low-power operation, infinite endurance, large on-off ratios, excellent write-error rates, nanosecond writing time, sub-nanosecond reading time, and good scalability. This project uses an interdisciplinary approach spanning materials, devices, circuits and architectures to realize such a memory and paradigm-shifting in-memory-processing architectures. The outcome will be durable, energy-efficient, pausable processing in polymorphic memories (DEEP3M), where computational capabilities are pushed directly into the high-capacity memories enabling massively parallel computation with fast and energy-efficient memory access. This approach builds on recent breakthroughs in physics of magnetic switching and advanced materials, and enables a transformative, holistic exploration of processing and memory by re-imaging the memory device as a computing element itself. This view will provide new insights and an entirely new paradigm for the semiconductor industry in the emerging era of Big Data. The team will also provide interdisciplinary educational opportunities to students and public alike.
View original record on NSF Award Search →