English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Recent works demonstrated the promise of using resistive random access memory(ReRAM) as an emerging technology to perform inherently parallel analog domainin-situ matrix-vector multiplication -- the intensive and key computation inDNNs. With weights stored in the ReRAM crossbar cells as conductance, when theinput vector is applied to word lines, the matrix-vector multiplication resultscan be generated as the current in bit lines. A key problem is that the weightcan be either positive or negative, but the in-situ computation assumes allcells on each crossbar column with the same sign. The current architectureseither use two ReRAM crossbars for positive and negative weights, or add anoffset to weights so that all values become positive. Neither solution isideal: they either double the cost of crossbars, or incur extra offsetcircuity. To better solve this problem, this paper proposes FORMS, afine-grained ReRAM-based DNN accelerator with polarized weights. Instead oftrying to represent the positive/negative weights, our key design principle isto enforce exactly what is assumed in the in-situ computation -- ensuring thatall weights in the same column of a crossbar have the same sign. It naturallyavoids the cost of an additional crossbar. Such weights can be nicely generatedusing alternating direction method of multipliers (ADMM) regularizedoptimization, which can exactly enforce certain patterns in DNN weights. Toachieve high accuracy, we propose to use fine-grained sub-array columns, whichprovide a unique opportunity for input zero-skipping, significantly avoidingunnecessary computations. It also makes the hardware much easier to implement.Putting all together, with the same optimized models, FORMS achievessignificant throughput improvement and speed up in frame per second over ISAACwith similar area cost.

FORMS: Fine-grained Polarized ReRAM-based In-situ Computation for Mixed-signal DNN Accelerator