In-memory computing (IMC) shows great potential as a non-von Neumann computational paradigm that can improve energy efficiency. Our research into intelligent compute memories includes both in-memory and near-memory approaches to address the main challenges of current architectures. These challenges include slow communication between big data memories and processing, as well as the overhead of data memory accesses.
To overcome these bottlenecks, near-memory co-processors can provide distributed computing resources at the periphery of memory arrays, while in-memory solutions leverage the read mechanism of memory arrays and specialized logic to perform arithmetic operations efficiently.
Blade bit-line computing | |
in-SRAM computing | |
Near Memory Computing | |
Software-defined SIMD |