
InMemory Nearest Neighbor Search with FeFET MultiBit ContentAddressable Memories
Nearest neighbor (NN) search is an essential operation in many applicati...
read it

Bit Error Robustness for EnergyEfficient DNN Accelerators
Deep neural network (DNN) accelerators received considerable attention i...
read it

Braininspired Cognition in Next Generation Racetrack Memories
Hyperdimensional computing (HDC) is an emerging computational framework ...
read it

Hypervector Design for Efficient Hyperdimensional Computing on Edge Devices
Hyperdimensional computing (HDC) has emerged as a new lightweight learn...
read it

Accelerating Bulk BitWise X(N)OR Operation in ProcessinginDRAM Platform
With VonNeumann computing architectures struggling to address computati...
read it

Bit Parallel 6T SRAM Inmemory Computing with Reconfigurable BitPrecision
This paper presents 6T SRAM cellbased bitparallel inmemory computing ...
read it

ERSFQ 8bit Parallel Binary Shifter for EnergyEfficient Superconducting CPU
We have designed and tested a parallel 8bit ERSFQ binary shifter that i...
read it
MIMHD: Accurate and Efficient Hyperdimensional Inference Using MultiBit InMemory Computing
Hyperdimensional Computing (HDC) is an emerging computational framework that mimics important brain functions by operating over highdimensional vectors, called hypervectors (HVs). Inmemory computing implementations of HDC are desirable since they can significantly reduce data transfer overheads. All existing inmemory HDC platforms consider binary HVs where each dimension is represented with a single bit. However, utilizing multibit HVs allows HDC to achieve acceptable accuracies in lower dimensions which in turn leads to higher energy efficiencies. Thus, we propose a highly accurate and efficient multibit inmemory HDC inference platform called MIMHD. MIMHD supports multibit operations using ferroelectric fieldeffect transistor (FeFET) crossbar arrays for multiplyandadd and FeFET multibit contentaddressable memories for associative search. We also introduce a novel hardwareaware retraining framework (HWART) that trains the HDC model to learn to work with MIMHD. For six popular datasets and 4000 dimension HVs, MIMHD using 3bit (2bit) precision HVs achieves (i) average accuracies of 92.6 (4.8 improvement over a GPU, and (iii) 38.4x (34.3x) speedup over a GPU, respectively. The 3bit × is 4.3x and 13x faster and more energyefficient than binary HDC accelerators while achieving similar accuracies.
READ FULL TEXT
Comments
There are no comments yet.