31 October 2022 in IEEE Micro
Deep neural networks (DNNs) are penetrating into a broad spectrum of applications and replacing manual algorithmic implementations, including the radio frequency communications domain with classical signal processing algorithms. However, the high throughput (gigasamples per second) and low latency requirements of this application domain pose a significant hurdle for adopting computationally demanding DNNs. In this article, we explore highly specialized DNN inference accelerator approaches on field-programmable gate arrays (FPGAs) for RadioML modulation classification. Using an automated end-to-end flow for the generation of the FPGA solution, we can easily explore a spectrum of solutions that optimize for different design targets, including accuracy, power efficiency, resources, throughput, and latency. By leveraging reduced precision arithmetic and customized streaming dataflow, we demonstrate a solution that meets the application requirements and outperforms alternative FPGA efforts by 3.5× in terms of throughput. Against modern embedded graphics processing units (GPUs), we measure $>\mathsf{10}\times$>10× higher throughput and $>\mathsf{100}\times$>100× lower latency under comparable accuracy and power envelopes.
27 May 2026 in ACM Transactions on Reconfigurable Technology and Systems
While neural network quantization effectively reduces the cost of matrix multiplications, aggressive quantization can expose non-matrix-multiply operations as significant performance and resource bottlenecks on embedded systems. Addressing such bottlenecks requires a comprehensive approach to tailoring the precision across operations in the inference computation. To this end, we introduce scaled-integer range analysis ( SIRA ), a static analysis technique employing interval arithmetic to determine the range, scale, and bias for tensors in quantized neural networks. We show how this information can be exploited to reduce the resource footprint of FPGA dataflow neural network accelerators via tailored bitwidth adaptation for accumulators and downstream operations, aggregation of scales and biases, and conversion of consecutive elementwise operations to thresholding operations. We integrate SIRA -driven optimizations into the open source FINN framework, then evaluate their effectiveness across a range of quantized neural network workloads and compare implementation alternatives for non-matrix-multiply operations. We demonstrate an average reduction of 17% for LUTs, 66% for DSPs, and 22% for accumulator bitwidths with SIRA optimizations, providing detailed benchmark analysis and analytical models to guide the implementation style for non-matrix layers. Finally, we open source SIRA to facilitate community exploration of its benefits across various applications and hardware platforms.