Non-linear functions are prevalent in Transformers and their lightweight variants, incurring substantial and frequently underestimated hardware costs. Previous state-of-the-art works optimize these operations by piece-wise linear approximation and store the parameters in look-up tables (LUT), but most of them require unfriendly high-precision arithmetics such as FP/INT 32 and lack consideration of integer-only INT quantization. This paper proposed a genetic LUT-Approximation algorithm namely GQA-LUT that can automatically determine the parameters with quantization awareness. The results demonstrate that GQA-LUT achieves negligible degradation on the challenging semantic segmentation task for both vanilla and linear Transformer models. Besides, proposed GQA-LUT enables the employment of INT8-based LUT-Approximation that achieves an area savings of 81.3~81.7% and a power reduction of 79.3~80.2% compared to the high-precision FP/INT 32 alternatives. Code is available at https://github.com/PingchengDong/GQA-LUT.
16 August 2022 in IEEE Transactions on Circuits and Systems II: Express Briefs
Much effort overlaps in designing different hardware to implement different Simultaneous localization and mapping (SLAM) algorithms. In this brief, a reconfigurable architecture with dedicated instruction sets allows the coprocessor to satisfy the pose estimation of a sample class of the SLAM algorithms, feature-based or learning-based methods, which can be decomposed to basic common operations. Furthermore, a memory-reused strategy in instructions was designed to avoid the demand for temporary memory for complex operations. Finally, two parallel computing cores are implemented to perform matrix operations and special computation about the pose estimation in floating-point and fixed-point arithmetic. These contribute to the low hardware resource usage and memory requirement, as illustrated in the experimental results.
With the wide application of autonomous mobile robots (AMRs), the visual inertial odometer (VIO) system that realizes the positioning function through the integration of a camera and inertial measurement unit (IMU) has developed rapidly, but it is still limited by the high complexity of the algorithm, the long development cycle of the dedicated accelerator, and the low power supply capacity of AMRs. This work designs a reconfigurable accelerated core that supports different VIO algorithms and has high area and energy efficiency, precision, and speed processing characteristics. Experimental results show that the loss of accuracy of the proposed accelerator is negligible on the most authoritative dataset. The on-chip memory usage of 70 KB is at least 10× smaller than the state-of-the-art works. Thus, the FPGA implementation’s hardware-resource consumption, power dissipation, and synthesis in the 28 nm CMOS outperform the previous works with the same platform.
15 December 2025 in IEEE Transactions on Circuits and Systems I: Regular Papers
Generating high-quality depth maps with accurate values is a critical research topic, and various depth estimation methods, such as Time-of-Flight (ToF) and Monocular Depth Estimation (MDE), are advancing rapidly. However, each single-sensor approach is inherently limited by the characteristics of its respective modality. Meanwhile, fused approaches using multiple sensors or dedicated trained models are often plagued by system complexity and limited generalization. In this paper, we propose an adaptive depth processing system based on the foundation transformer and ToF fusion, aiming to harness the precision of ToF data and the high-quality depth priors provided by the foundation model in a monocular and zero-shot manner. A hardware–algorithm co-design partitions the computation between a dedicated ASIC for compute-intensive foundation transformer acceleration and an MPSoC FPGA for potential reconfigurable fusion, yielding 36.4 ms latency and 1.9 W power under a 27.6 GOPS/frame workload. Zero-shot evaluations on various ToF data sources including FLAT, TICaM, and in-house captures confirm strong cross-domain generalization, demonstrating the system’s adaptability, efficiency, and accuracy.
1 January 2026 in IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems
SRAM computing-in-memory (CIM) accelerators have evolved as a promising solution to the memory wall problem in neural network (NN) models. By integrating memory and compute resources in each macro, CIM accelerators offer massive in-situ computing parallelism and large memory capacity, enabling spatial mapping with layer fusion and potentially keeping layers stationary in CIM. However, CIM’s memory-compute coupling (MCC) feature poses challenges in designing CIM accelerators. From an architecture aspect, designers must balance CIM’s memory and compute resources by optimizing the macro’s memory-compute ratio (MCR) configuration across diverse scenarios. From a mapping aspect, conventional mappings, which allocate each macro exclusively to each layer, face two major problems: 1) a layer-fusion dilemma (the accelerator suffers from excessive memory access due to layer replications or performance degradation due to load imbalance) and 2) a layer-eviction issue (storing layers stationary in CIM is usually infeasible due to limited CIM capacity). To address these challenges, this article introduces MCC-DSE, an MCC-aware Design Space Exploration framework for architecture-mapping co-optimization of CIM accelerators. We also propose a three-axis CIM division mapping, which interleaves multiple layers in each macro to concurrently optimize memory access and performance during layer fusion as well as reserves a part of CIM memory in each macro for layer pinning. Compared to baseline architecture and mapping, MCC-DSE shows a $1.4\times $ –$8.3\times $ EDP reduction across various workloads and chip areas.