Arun Subramaniyan

Arun Subramaniyan

Auto-generated profile, based on publicly available information

Arun Subramaniyan

Department of Electrical Engineering and Computer Science, University of Michigan, Ann Arbor, MI 48109 USA

Research Focus & Metrics

Academic Career Length

15Years (2012-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
23 September 2019 in IEEE Journal of Solid-State Circuits
Verified
Journal Article
Closed access
JWJingcheng Wang
Jingcheng Wang
XWXiaowei Wang
Xiaowei Wang
CECharles Eckert
Charles Eckert
ASArun Subramaniyan
Arun Subramaniyan
RDReetuparna Das
Reetuparna Das
DSDennis Sylvester
Dennis Sylvester
This article proposes a general-purpose hybrid in-/near-memory compute SRAM (CRAM) that combines an 8T transposable bit cell with vector-based, bit-serial in-memory arithmetic to accommodate a wide range of bit-widths, from single to 32 or 64 bits, as well as a complete set of operation types, including integer and floating-point addition, multiplication, and division. This approach provides the flexibility and programmability necessary for evolving software algorithms ranging from neural networks to graph and signal processing. The proposed design was implemented in a small Internet of Things (IoT) processor in the 28-nm CMOS consisting of a Cortex-M0 CPU and 8 CRAM banks of 16 kB each (128 kB total). The system achieves 475-MHz operation at 1.1 V and, with all CRAMs active, produces 30 GOPS or 1.4 GFLOPS on 32-bit operands. It achieves an energy efficiency of 0.56 TOPS/W for 8-bit multiplication and 5.27 TOPS/W for 8-bit addition at 0.6 V and 114 MHz.

17 June 2023
Verified
Conference Paper
Open access
YGYufeng Gu
Yufeng Gu
ASArun Subramaniyan
Arun Subramaniyan
TDTim Dunn
Tim Dunn
AKAlireza Khadem
Alireza Khadem
KCKuan-Yu Chen
Kuan-Yu Chen
RDReetuparna Das
Reetuparna Das
Genomics is playing an important role in transforming healthcare. Genetic data, however, is being produced at a rate that far outpaces Moore's Law. Many efforts have been made to accelerate genomics kernels on modern commodity hardware such as CPUs and GPUs, as well as custom accelerators (ASICs) for specific genomics kernels. While ASICs provide higher performance and energy efficiency than general-purpose hardware, they incur a high hardware design cost. Moreover, in order to extract the best performance, ASICs tend to have significantly different architectures for different kernels. The divergence of ASIC designs makes it difficult to run commonly used modern sequencing analysis pipelines due to software integration and programming challenges. With the observation that many genomics kernels are dominated by dynamic programming (DP) algorithms, this paper presents GenDP, a framework of dynamic programming acceleration including DPAx, a DP accelerator, and DPMap, a graph partitioning algorithm that maps DP objective functions to the accelerator. DPAx supports DP kernels with various dependency patterns, such as 1D and 2D DP tables and long-range dependencies in the graph structure. DPAx also supports different DP objective functions and precisions required for genomics applications. GenDP is evaluated on genomics kernels in both short-read and long-read analysis pipelines, achieving 157.8× throughput/mm2 over GPU baselines and 132.0× throughput/mm2 over CPU baselines.

16 December 2020 in IEEE Solid-State Circuits Letters
Verified
Journal Article
Closed access
XWXiao Wu
Xiao Wu
ASArun Subramaniyan
Arun Subramaniyan
ZWZhehong Wang
Zhehong Wang
SNSatish Narayanasamy
Satish Narayanasamy
RDReetuparna Das
Reetuparna Das
DBDavid Blaauw
David Blaauw
We present the first ASIC accelerator for a pair-Hidden-Markov-Model (Pair-HMM) in DNA variant calling, which conventionally requires 250T FLOPs per sequenced human genome. Using a hardware-algorithm co-design, we opportunistically replace floating point (FP) multiplication with 20b log-domain addition while employing bound checks to maintain (provable) correct results in downstream processing. FP computation is reduced by 43× on real human genome data. Implemented in 40nm CMOS, the 5.67 mm2 accelerator demonstrates 17.3G cell updates per second (CUPS) throughput, marking a 6.6× improvement over our baseline ASIC implementation and 355× GCUPS/mm2 improvement over an FPGA implementation 2.

12 October 2022 in IEEE Transactions on Computers
Verified
Journal Article
Closed access
CECharles Eckert
Charles Eckert
ASArun Subramaniyan
Arun Subramaniyan
XWXiaowei Wang
Xiaowei Wang
CACharles Augustine
Charles Augustine
RIRavishankar Iyer
Ravishankar Iyer
RDReetuparna Das
Reetuparna Das
This paper presents the Eidetic architecture, which is an SRAM-based ASIC neural network accelerator that eliminates the need to continuously load weights from off-chip, while also minimizing the need to go off chip for intermediate results. Using in-situ arithmetic in the SRAM arrays, this architecture can supports a variety of precision types allowing for effective inference. We also present different data mapping policies for matrix-vector based networks (RNN and MLP) on the Eidetic architecture and describe the tradeoffs involved. With this architecture, multiple layers of a network can be concurrently mapped, storing both the layer weights and intermediate results on-chip, removing the energy and latency penalty of off-chip memory accesses. We evaluate Eidetic on Google's Neural Machine Translation System (GNMT) encoder and demonstrate a 17.20× increase in throughput and 7.77× reduction in average latency over a single TPUv2 chip.

1 January 2015 in Journal of Aerospace Information Systems
Verified
Journal Article
Closed access
ASAnkur Srivastava
Ankur Srivastava
ASArun K. Subramaniyan
Arun K. Subramaniyan
Liping WangLiping Wang
Liping Wang
This paper presents a hybrid Bayesian solution to the multidisciplinary uncertainty quantification challenge posed by NASA Langley Research Center. The hybrid approach builds probabilistic models of the outputs by combining the simulations and “experimental” observations provided by NASA. The proposed method is a synergy between Gaussian process surrogate modeling and Bayesian principles, and the resulting tool allows construction of surrogate models, parameter calibration, and variance-based global sensitivity analysis. Besides Sobol indices, a cumulative-density-function-based sensitivity to the probability of failure is also explored. Particle swarm optimization has been used in conjunction with the Gaussian process surrogates to find extreme values of both the mean value of the worst-case requirement function and the failure probability. The capability of the proposed techniques is demonstrated on the four subproblems of uncertainty characterization, sensitivity analysis, uncertainty propagation, and ...

22 September 2020 in IEEE Journal of Solid-State Circuits
Verified
Journal Article
Closed access
ZWZhehong Wang
Zhehong Wang
TZTianjun Zhang
Tianjun Zhang
DFDaichi Fujiki
Daichi Fujiki
ASArun Subramaniyan
Arun Subramaniyan
XWXiao Wu
Xiao Wu
DBDavid Blaauw
David Blaauw
Advances in DNA-sequencing technology have far outpaced Moore's law, imposing significant challenges on the computationally intensive secondary analysis in the sequencing pipeline. An accelerator for seed extension, a critical and computationally intensive step in genome sequencing, is proposed. The accelerator, implementing a string-independent automata, consists of a triangular array of 25x 2 custom-designed processing elements. It performs 2.46 million reads per second (MRPS), achieving a 1581x improvement in power efficiency and 165.5x smaller silicon footprint compared to a system setting with dual-socket Xeon E5-2697 v3 server processors.