Energy-Efficient and Scalable Access Scheme for Non-Volatile Memory Devices
Published:8/18/2021Description:
The Problem:
CMOS memory technology is plagued by scaling issues such as oxide aging, capacitive coupling, and cell-to-cell interference. Non-volatile memory (NVM) devices like memory resistors (memristors) offer increased power efficiency and reduced chip area, allowing them to emerge as a viable alternative. As more and more NVM devices are used in numerous applications, there is need for a systematic approach to programming or reading to and from these devices.
The Solution:
Researchers at the University of Tennessee have devised a system-level scheme to configure or extract the multi-bit information from NVM devices. The distributed NVM elements are sequentially configured or programmed by scanning memory states into temporary scan registers. These temporary values are then used to write to NVM elements within a corresponding memory bank. The scan registers can also be used to read out data bits from the non-volatile memory devices. To facilitate read and write operations, two scan registers are used interchangeably to separate and parallelize the programming/reading and shifting in/out of data bits.
Benefits:
- Efficient parallelization of scanning of configuration bits and accessing the NVM devices for reading or writing
- Size or length of a scan register does not on the number of NVM devices
- Scalable to account for greater number of discrete programmable states of the NVM devices
- Scalable to accommodate a high number of NVM devices with limited peripheral circuitry
- Estimated energy usage is only 3.48 pJ/bit, compared to 24.41 pJ/bit for multi-bit programming schemes
The Inventors:
Dr. Garrett Rose is an associate professor in the Department of Electrical Engineering and Computer Science at UT. He received his Ph.D. degree in electrical engineering from the University of Virginia in 2006. His research interests include nanoelectric circuit design and computer architectures, memristors and memristive systems, and neuromorphic computing.
Musabbir Adnan is a graduate student in the Department of Electrical Engineering and Computer Science at UT. His research interests include neuromorphic computing, VLSI design of nanoelectronic circuits, and emerging devices.