DAC 2021

121 papers
The 2021 Design Automation Conference (DAC) featured 151 papers focusing heavily on the intersection of hardware design and machine learning. A primary research direction was Neural Architecture Search (NAS), specifically targeting hardware-aware, memory-aware, and heterogeneous layer-wise configurations. Significant attention was also given to Spiking Neural Networks (SNNs), with studies exploring energy-efficient training, conversion, and time-to-first-spike coding. Other prominent themes included edge intelligence, context-adaptive computation offloading, and the performance bottlenecks in high-speed autonomous systems. Notable technical contributions included the RASA systolic array matrix engine, the Gemmini full-stack integration for deep-learning evaluation, and ReRAM-based acceleration for Graph Convolutional Networks (PIMGCN). Security and reliability were addressed through works on Deep Neural Network integrity (AID) and trusted execution environments for CPU-FPGA architectures (SGX-FPGA). Research also explored aggressive voltage scaling and federated learning via the Helios framework. Prolific contributors to this edition included Mohsen Imani, Vijay Janapa Reddi, Muhammad Shafique, and Deliang Fan.

Top Research Directions

Research directions most represented by DAC 2021 papers.

Robust Deep Learning on Noisy Analog Memory Crossbars 3
Hardware-Aware Neural Architecture Search 3
Automated Hardware Design Space Exploration 2
Securing Integrated Circuits Against Untrusted Foundries 2
Cross Layer Co Design for Deep Learning Accelerators 2
Closing the Neuromorphic Hardware Deployment Gap 2
Integrating Physical Awareness into Logic Synthesis 1
Fine-Grained and Dynamic Memory Layout Randomization 1
In-Situ Defect Detection for Real-Time Quality Control 1
Hybrid Hardware Architectures for Graph Neural Network Acceleration 1
The Non-Linear Operation Bottleneck in Quantized Transformers 1
Hardware and Computational Bottlenecks of Complex Activations 1
Framework Execution Overhead in Deep Learning 1
Partial Connections and Path Binarization for Architecture Search 1
Automated Tuning of Hardware Execution Schedules 1

Top Authors

AuthorPapers
Mohsen Imani5
Yiyu Shi4
Cheng Liu3
David Z. Pan3
Deliang Fan3
Huawei Li3
Muhammad Shafique3
Vijay Janapa Reddi3
Xiaowei Li3
Alejandro Hernández-Cano2
Bei Yu2
Caiwen Ding2
Chaojian Li2
CUNXI YU2
Deming Chen2
Guojie Luo2
Haowen Fang2
Jinho Lee2
Jörg Henkel2
Kenli Li2

Impactful Papers

Papers from DAC 2021 that shaped research clusters.

Influential

Papers

Title
Safety in Autonomous Driving: Can Tools Offer Guarantees?
SEALing Neural Network Models in Encrypted Deep Learning Accelerators
SGL: Spectral Graph Learning from Measurements
SGX-FPGA: Trusted Execution Environment for CPU-FPGA Heterogeneous Architecture
Skew-Oblivious Data Routing for Data Intensive Applications on FPGAs with HLS
SLAP: A Supervised Learning Approach for Priority Cuts Technology Mapping
SmartBoost: Lightweight ML-Driven Boosting for Thermally-Constrained Many-Core Processors
Softermax: Hardware/Software Co-Design of an Efficient Softmax for Transformers
SparkXD: A Framework for Resilient and Energy-Efficient Spiking Neural Network Inference using Approximate DRAM
SpikeDyn: A Framework for Energy-Efficient Spiking Neural Networks with Continual and Unsupervised Learning Capabilities in Dynamic Environments
SpV8: Pursuing Optimal Vectorization and Regular Computation Pattern in SpMV
Statheros: Compiler for Efficient Low-Precision Probabilistic Programming
StocHD: Stochastic Hyperdimensional System for Efficient and Robust Learning from Raw Data
Synthesizing Barrier Certificates of Neural Network Controlled Continuous Systems via Approximations
TARe: Task-Adaptive in-situ ReRAM Computing for Graph Learning
TinyML: Current Progress, Research Challenges, and Future Roadmap
Towards Improving the Trustworthiness of Hardware based Malware Detector using Online Uncertainty Estimation
Towards Resilient Deployment of In-Memory Neural Networks with High Throughput
Training Acceleration for Deep Neural Networks: A Hybrid Parallelization Strategy
Ultrafast CPU/GPU Kernels for Density Accumulation in Placement
Universal Symmetry Constraint Extraction for Analog and Mixed-Signal Circuits with Graph Neural Networks

« prev Page 3/3