SPSubhankar Pal

Subhankar Pal

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article

Previous Organizations / Institutes

The organizations / institutes from the scholar's other articles in the past ten years


Subhankar Pal has a total of 68 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Dramaton: A Near-DRAM Accelerator for Large Number Theoretic Transforms

    by Yongmo Park, Subhankar Pal, Aporva Amarnath, Karthik Swaminathan, Wei D. Lu, Alper Buyuktosunoglu, Pradip Bose

    With the rising popularity of post-quantum cryptographic schemes, realizing practical implementations for realworld applications is still a major challenge. A major bottleneck in such schemes is the fetching and processing of large polynomials in the Number Theoretic Transform (NTT), which makes non Von Neumann paradigms, such as near-memory processing, a viable option. We, therefore, propose a novel near-DRAM NTT accelerator design, called DRAMATON. Additionally, we introduce a conflict-free mapping algorithm that enables DRAMATON to process large NTTs with minimal hardware overhead using a fixed-permutation network. DRAMATON achieves 5-207× speedup in latency over the state-of-the-art and 97× improvement in EDP over a recent near-memory NTT accelerator.


  • Efficient Pruning for Machine Learning Under Homomorphic Encryption

    by Ehud Aharoni, Moran Baruch, Pradip Bose, Alper Buyuktosunoglu, Nir Drucker, Subhankar Pal, Tomer Pelleg, Kanthi Sarpatwar, Hayim Shaul, Omri Soceanu, et al.


  • Open Access

    OnSRAM: Efficient Inter-Node On-Chip Scratchpad Management in Deep Learning Accelerators

    by Subhankar Pal, Swagath Venkataramani, Viji Srinivasan, Kailash Gopalakrishnan

    Hardware acceleration of Artificial Intelligence (AI) workloads has gained widespread popularity with its potential to deliver unprecedented performance and efficiency. An important challenge remains in how AI accelerators are programmed to sustain high utilization without impacting end-user productivity. Prior software optimizations start with an input graph and focus on node-level optimizations, viz. dataflows and hierarchical tiling, and graph-level optimizations such as operation fusion. However, little effort has been devoted to inter-node on-chip scratchpad memory (SPM) management in Deep Learning (DL) accelerators, whose significance is bolstered by the recent trends in complex network topologies and the emergence of eager execution in DL frameworks. We characterize and show that there exists up to a 5.2× performance gap in DL inference to be bridged using SPM management and propose OnSRAM, a novel SPM management framework integrated with the compiler runtime of a DL accelerator. We develop two variants, viz. OnSRAM-Static, which works on static graphs to identify data structures that can be lucratively held on-chip based on their size, liveness and significance, and OnSRAM-Eager, which targets an eager execution model (no graph) and uses a history-based speculative scheme to hold/discard data structures. We integrate OnSRAM with TensorFlow and analyze it on multiple accelerator configurations. Across a suite of 12 images, objects, and language networks, on a 3 TFLOP system with a 2 MB SPM and 32 GBps external memory bandwidth, OnSRAM-Static and OnSRAM-Eager achieve 1.02–4.8× and 1.02–3.1× reduction in inference latency (batch size of 1), over a baseline with no SPM management. In terms of energy savings, we observe average reductions of 1.51× (up to 4.1×) and 1.23× (up to 2.9×) for the static and eager execution scenarios, respectively.


  • Versa: A 36-Core Systolic Multiprocessor With Dynamically Reconfigurable Interconnect and Memory

    by Sung Kim, Morteza Fayazi, Alhad Daftardar, Kuan-Yu Chen, Jielun Tan, Subhankar Pal, Tutu Ajayi, Yan Xiong, Trevor Mudge, Chaitali Chakrabarti, et al.

    We present Versa, an energy-efficient 36-core systolic multiprocessor with dynamically reconfigurable interconnects and memory. Versa leverages reconfigurable functional units and systolic-enhanced ARM cores to adapt for different algorithm characteristics, providing optimized bandwidth, access latency, and data reuse. Hardware support for crucial thread-synchronization operations enables a tree-based algorithm with 6.5 $\times $ improvement in synchronization latency. Measured on a diverse set of compute kernels, Versa’s design features culminate in median energy-efficiency improvements of 37.2 $\times $ and 11.6 $\times $ over mobile CPU and GPU baselines, respectively.


  • Heterogeneity-Aware Scheduling on SoCs for Autonomous Vehicles

    by Aporva Amarnath, Subhankar Pal, Hiwot Tadese Kassa, Augusto Vega, Alper Buyuktosunoglu, Hubertus Franke, John-David Wellman, Ronald Dreslinski, Pradip Bose

    Fully autonomous vehicles (AVs) must meet stringent real-time performance and safety-criticality constraints of multiple applications simultaneously in highly dynamic environmental conditions. To enable such a system, carefully selected accelerators and general-purpose cores on Systems-on-Chips (SoCs) are required. However, schedulers that are agnostic to this heterogeneity lead to inefficient utilization of hardware resources and increase the time to complete the AV's mission. As a solution, our letter proposes a heterogeneity-aware, multi-level scheduler called HetSched. HetSched leverages run-time information about the underlying heterogeneous SoC, along with the applications’ real-time constraints to improve an AV's mission time. Our evaluation demonstrates a reduction in mission time of 2.0-4.8×, when compared against state-of-the-art (SOTA) schedulers. Furthermore, when used as part of an SoC design space exploration loop, in comparison to the SOTA schedulers, HetSched reduces the number of accelerators required by an SoC to safely complete the AV's mission by 1.9× on average.


  • Accelerating Deep Neural Network Computation on a Low Power Reconfigurable Architecture

    by Y. Xiong, J. Zhou, S. Pal, D. Blaauw, H. S. Kim, T. Mudge, R. Dreslinski, C. Chakrabarti

    Recent work on neural network architectures has focused on bridging the gap between performance/efficiency and programmability. We consider implementations of three popular neural networks, ResNet, AlexNet and ASGD weight-dropped Recurrent Neural Network (AWD RNN) on a low power programmable architecture, Transformer. The architecture consists of light-weight cores interconnected by caches and crossbars that support run-time reconfiguration between shared and private cache mode operations. We present efficient implementations of key neural network kernels and evaluate the performance of each kernel when operating in different cache modes. The best-performing cache modes are then used in the implementation of the end-to-end network. Simulation results show superior performance with ResNet, AlexNet and AWD RNN achieving 188.19 GOPS/W, 150.53 GOPS/W and 120.68 GOPS/W, respectively, in the 14 nm technology node.


  • Open Access

    A 7.3 M Output Non-Zeros/J, 11.7 M Output Non-Zeros/GB Reconfigurable Sparse Matrix–Matrix Multiplication Accelerator

    by Dong-Hyeon Park, Subhankar Pal, Siying Feng, Paul Gao, Jielun Tan, Austin Rovinski, Shaolin Xie, Chun Zhao, Aporva Amarnath, Timothy Wesley, et al.

    A sparse matrix-matrix multiplication (SpMM) accelerator with 48 heterogeneous cores and a reconfigurable memory hierarchy is fabricated in 40-nm CMOS. The compute fabric consists of dedicated floating-point multiplication units, and general-purpose Arm Cortex-M0 and Cortex-M4 cores. The on-chip memory reconfigures scratchpad or cache, depending on the phase of the algorithm. The memory and compute units are interconnected with synthesizable coalescing crossbars for efficient memory access. The 2.0-mm x 2.6-mm chip exhibits 12.6x (8.4x) energy efficiency gain, 11.7x (77.6x) off-chip bandwidth efficiency gain, and 17.1x (36.9x) compute density gains against a high-end CPU (GPU) across a diverse set of synthetic and real-world power-law graph-based sparse matrices.


  • A 7.3 M Output Non-Zeros/J Sparse Matrix-Matrix Multiplication Accelerator using Memory Reconfiguration in 40 nm

    by Subhankar Pal, Dong-Hyeon Park, Siying Feng, Paul Gao, Jielun Tan, Austin Rovinski, Shaolin Xie, Chun Zhao, Aporva Amarnath, Timothy Wesley, et al.


  • A 7.3 M Output Non-Zeros/J Sparse Matrix-Matrix Multiplication Accelerator using Memory Reconfiguration in 40 nm

    by Subhankar Pal, Dong-Hyeon Park, Siying Feng, Paul Gao, Jielun Tan, Austin Rovinski, Shaolin Xie, Chun Zhao, Aporva Amarnath, Timothy Wesley, et al.


  • Parallelism Analysis of Prominent Desktop Applications: An 18- Year Perspective

    by Siying Feng, Subhankar Pal, Yichen Yang, Ronald G. Dreslinski

    Improvements in clock speed and exploitation of Instruction-Level Parallelism (ILP) hit a roadblock during mid-2000s. This, coupled with the demise of Dennard scaling, led to the rise of multi-core machines. Today, multi-core processors are ubiquitous and architects have moved to specialization to work around the walls hit by single-core performance and chip Thermal Design Power (TDP). The pressure of innovation in the aftermath of Dennard scaling is shifting to software developers, who are required to write programs that make the most effective use of underlying hardware. This work presents quantitative and qualitative analyses of how software has evolved to reap the benefits of multi-core and heterogeneous computers, compared to state-of-the-art systems in 2000 and 2010. We study a wide spectrum of commonly-used applications on a state-of-the-art desktop machine and analyze two important metrics, Thread-Level Parallelism (TLP) and GPU utilization. We compare the results to prior work over the last two decades, which state that 2-3 CPU cores are sufficient for most applications and that the GPU is usually under-utilized. Our analyses show that the harnessed parallelism has improved and emerging workloads show good utilization of hardware resources. The average TLP across the applications we study is 3.1, with most applications attaining the maximum instantaneous TLP of 12 during execution. The GPU is over-provisioned for most applications, but workloads such as cryptocurrency mining utilize it to the fullest. Overall, we conclude that the effectiveness of software in utilizing the underlying hardware has improved, but still has scope for optimizations.