Guoyu Li

Guoyu Li, Yang Cao, Lucas H L Ng et al.

With network requirements diverging across emerging applications, latency-critical services demand minimal logic delay, while hyperscale training and collectives require sustained line-rate throughput for synchronized bulk transfers. This divergence creates an urgent need for custom network switches tailored to specialized protocols and application-specific traffic patterns. This paper presents SPAC (Switch and Protocol Adaptive Customization), a novel approach that automates the generation of FPGA-based network switches co-optimized for custom protocols and application-specific traffic patterns. SPAC introduces a unified workflow with a domain-specific language (DSL) for protocol-architecture co-design, a library of modular HLS-based adaptive switch components, and a trace-aware Design Space Exploration (DSE) engine. By providing a multi-fidelity simulation stack, SPAC enables rapid identification of Pareto-optimal designs prior to deployment. We demonstrate the efficacy of the domain-specific adaptation of SPAC across a spectrum of real-world scenarios, spanning from latency-sensitive sensor and HFT networks to hyperscale datacenter fabrics. Experimental results show that by tailoring the micro-architecture and protocol to the specific workload, SPAC-generated designs reduce LUT and BRAM usage by 55% and 53%, respectively. Compared to fixed-architecture counterparts, SPAC delivers latency reductions ranging from 7.8% to 38.4% across various tasks while maintaining adequate resource consumption and packet drop rate.

Zhiwen Mo, Guoyu Li, Hao et al.

Advances in hybrid bonding and packaging have driven growing interest in 3D DRAM-stacked accelerators with higher memory bandwidth and capacity. As LLMs scale to hundreds of billions or trillions of parameters, distributed inference across multiple 3D chips becomes essential. With cross-stack co-design increasingly critical, we propose DeepStack, an accurate and efficient performance model and tool to enable early-stage system-hardware co-design space exploration (DSE) for distributed 3D-stacked AI systems. At the hardware level, DeepStack captures fine-grained 3D memory semantics such as transaction-aware bandwidth, bank activation constraints, buffering limitations, and thermal-power modeling. At the system level, DeepStack incorporates comprehensive parallelization strategies and execution scheduling for distributed LLM inference. With novel modeling techniques such as dual-stage network abstraction and tile-level compute-communication overlap, we achieve up to 100,000x faster runtime over state-of-the-art simulators at comparable accuracy, cross-validated against our in-house 3D designs, NS-3 backend (2.12%), and vLLM serving on 8xB200 GPUs (12.18%). With hierarchical design space search, DeepStack enables efficient exploration over 2.5x10^14 design points spanning 3D-stacked DRAM layers, DRAM vertical connectivity, interconnect, compute-memory allocation, and distributed scheduling. Compared with baseline designs, DeepStack achieves up to 9.5x higher throughput through co-optimized parallelism and 3D architecture search. Our DSE further reveals that batch size drives a more fundamental architectural divide than the prefill/decode distinction, and that parallelism strategy and hardware architecture are tightly coupled -- incomplete schedule search leads to permanently suboptimal silicon irrecoverable by software tuning. We intend to open source DeepStack to support future research.

Guoyu Li, Pengbo Zheng, Jian Weng et al.

Convolutional Neural Networks (CNNs) remain prevalent in computer vision applications, and FPGAs, known for their flexibility and energy efficiency, have become essential components in heterogeneous acceleration systems. However, traditional FPGAs face challenges in balancing performance and versatility due to limited on-chip resources. AMD's Versal ACAP architecture, tailored for AI applications, incorporates AI Engines (AIEs) to deliver high computational power. Nevertheless, the platform suffers from insufficient memory bandwidth, hindering the full utilization of the AIEs' theoretical performance. In this paper, we present DPUV4E for the Versal architecture, providing configurations ranging from 2PE ($32.6$ TOPS) to 8PE ($131.0$ TOPS). We design two computation units, Conv PE and DWC PE, to support different computational patterns. Each computation unit's data flow efficiently utilizes the data reuse opportunities to mitigate bandwidth bottlenecks. Additionally, we extend the functionality of each PE to utilize AIEs for non-convolutional operations, reducing resource overhead. Experiments on over 50 models show that compared to previous designs, our design provides $8.6\times$ the TOPS/W of traditional FPGA-based DPU designs, while reducing DSP usage by $95.8\%$, LUT usage by $44.7\%$, and latency to $68.5\%$ under single-batch conditions. For end-to-end inference, our design improving throughput by up to $2.2\times$ for depth-wise convolution models and up to $1.3\times$ for standard models.