Search

As Graph Convolutional Networks (GCNs) have emerged as a promising solution for graph representation learning, designing specialized GCN accelerators has become an important challenge. An analysis of GCN workloads shows that the main bottleneck of GCN processing is not computation but the memory latency of intensive off-chip data transfer. Therefore, minimizing off-chip data transfer is the primary challenge for designing an efficient GCN accelerator. To address this challenge, optimization is initialized by considering GCNs as tiled matrix multiplication. In this paper, we optimize off-chip memory access from both the in- and out-of-tile perspectives. From the out-of-tile perspective, we find optimal tile configurations of given datasets and on-chip buffer capacity, then observe the dataflow across phases and layers. Inter-layer phase fusion dataflow with optimal tile configuration reduces data transfer of intermediate outputs. From the in-tile perspective, due to the sparsity of tiles, tiles have redundant data which does not participate in computation. Redundant data load is eliminated with hardware support. Finally, we introduce an efficient GCN inference accelerator, EGCN, specialized for minimizing off-chip memory access. EGCN achieves 41.9% off-chip DRAM access reduction, 1.49× speedup, and 1.95× energy efficiency improvement on average over the state-of-the-art accelerators. [ABSTRACT FROM AUTHOR]

Emerging device-based digital processing-in-memory (PIM) architectures have been actively studied due to their energy and area efficiency derived from analog to digital converter (ADC)-less PIM hardware. However, digital PIM architectures generally need large extra memories to copy parameters, and they also suffer from low computation per memory-cycle efficiencies. In this paper, we present a novel spin-orbit torque magnetic random access memory (SOT-MRAM) based digital PIM architecture to alleviate the extra memory size burden and computation cycle issues. First, we propose the spintronics-assisted logic-in-memory (SLIM) cells to support efficient digital logic operations inside memories, where the voltage-controlled magnetic anisotropy (VCMA) is exploited to enhance the computation per memory-cycle efficiencies. In addition, crossed input source PIM (CRISP) architecture is proposed to extend the merits of SLIM cells by eliminating the extra memories for parameter copying while significantly improving the degree of parallel processing. An intra-memory pipelining scheme is also considered to further increase the throughput of CRISP. The proposed CRISP architecture has been implemented using 28 nm CMOS process, and it presents 1.10 TOPS/W and 0.95 TOPS/mm2, showing considerable improvements of energy efficiency and throughput per area, compared to the state-of-the-art digital PIM architecture. Finally, to evaluate the impact of computation errors induced from the SOT devices and circuits in CRISP architecture, classification accuracy simulations have been performed while applying computation errors. [ABSTRACT FROM AUTHOR]

Reed-Solomon (RS) codes are used in many storage systems for failure recovery. In popular software implementations, RS codes are defined by using a parity check matrix that is either a Cauchy matrix padded with an identity or a Vandermonde matrix. The encoding complexity can be reduced by searching for a Cauchy matrix that has a smaller number of ‘1's in its bit matrices or exploiting Reed-Muller (RM) transform in the Vandermonde matrix multiplication. This article proposes two new approaches that improve upon the previous schemes. In our first approach, different constructions of finite fields are explored to further reduce the number of ‘1's in the bit matrices of the Cauchy matrix and a new searching method is developed to find the matrices with minimum number of ‘1's. Our second approach defines RS codes using a parity check matrix in the format of a Vandermonde matrix concatenated with an identity matrix so that the multiplication with the inverse erasure columns in the encoding is eliminated and the decoding can be carried out using simplified formulas. The Vandermonde matrix in such an unconventional RS code definition needs to be constructed using finite field elements in non-consecutive order. A modification is also developed in this article to enable the application of the RM transform in this case to reduce the matrix multiplication complexity. For 4-erasure-correcting RS codes over $GF(2^8)$ G F (2 8) , the two proposed approaches increase the encoding throughput by 40 and 15 percent on average over the prior works based on Cauchy matrix and Vandermonde matrix with RM transform, respectively, for a range of codeword length. Moreover, the decoding throughput is also significantly improved. [ABSTRACT FROM AUTHOR]