The massive amounts of data generated by camera sensors motivate data processing inside pixel arrays, i.e., at the extreme-edge. Several critical developments have fueled recent interest in the processing-in-pixel-in-memory paradigm for a wide range of visual machine intelligence tasks, including (1) advances in 3D integration technology to enable complex processing inside each pixel in a 3D integrated manner while maintaining pixel density, (2) analog processing circuit techniques for massively parallel low-energy in-pixel computations, and (3) algorithmic techniques to mitigate non-idealities associated with analog processing through hardware-aware training schemes. This article presents a comprehensive technology-circuit-algorithm landscape that connects technology capabilities, circuit design strategies, and algorithmic optimizations to power, performance, area, bandwidth reduction, and application-level accuracy metrics. We present our results using a comprehensive co-design framework incorporating hardware and algorithmic optimizations for various complex real-life visual intelligence tasks mapped onto our P2M paradigm.
The exponential emergence of Field-Programmable Gate Arrays (FPGAs) has accelerated research on hardware implementation of Deep Neural Networks (DNNs). Among all DNN processors, domain-specific architectures such as Google’s Tensor Processor Unit (TPU) have outperformed conventional GPUs (Graphics Processing Units) and CPUs (Central Processing Units). However, implementing low-power TPUs in reconfigurable hardware remains a challenge in this field. Voltage scaling, a popular approach for energy savings, can be challenging in FPGAs, as it may lead to timing failures if not implemented appropriately. This work presents an ultra-low-power FPGA implementation of a TPU for edge applications. We divide the systolic array of a TPU into different FPGA partitions based on the minimum slack value of different design paths of Multiplier Accumulators (MACs). Each partition uses different near-threshold (NTC) biasing voltages to run its FPGA cores. The biasing voltage for each partition is roughly calculated by the proposed static schemes. However, further calibration of biasing voltage is performed by the proposed runtime scheme. To overcome the timing failure caused by NTC, the MACs with higher minimum slack are placed in lower-voltage partitions, while the MACs with lower minimum slack paths are placed in higher-voltage partitions. The proposed architecture is implemented in a commercial platform, namely with Xilinx FPGA and academic platform with 22 nm, 45 nm and 130 nm FPGAs. Any timing error caused by NTC can be caught by the Razor flipflop used in each MAC. The proposed voltage-scaled, partitioned systolic array can save 3.1% to 11.6% of dynamic power in and tools, respectively, depending on the FPGA technology, partition size, number of partitions and biasing voltages. The normalized performance and accuracy of benchmark models running on our low-power TPU are very competitive compared to existing literature.
Efficient and effective on-line detection and correction of bad pixels can improve yield and increase the expected lifetime of image sensors. This paper presents a comprehensive Deep Learning (DL) based on-line detection-correction approach, suitable for a wide range of pixel corruption rates. A confidence calibrated segmentation approach is introduced, which achieves nearly perfect bad pixel detection, even with few training samples. A computationally light-weight correction algorithm is proposed for low rates of pixel corruption, that surpasses the accuracy of traditional interpolation-based techniques. We also propose an autoencoder based image reconstruction approach which alleviates the need for prior bad pixel detection and yields promising results for high rates of pixel corruption. Unlike previous methods, which use proprietary images, we demonstrate the efficacy of the proposed methods on the open-source Samsung S7 ISP and MIT-Adobe FiveK datasets. Our approaches yield up to 99.6% detection accuracy with <0.6% false positives and corrected images within 1.5% average pixel error from 70% corrupted images.
Vision-Language Models (VLMs) have achieved remarkable progress in multimodal reasoning and generation, yet their high computational demands remain a major challenge. Diffusion Vision-Language Models (DVLMs) are particularly attractive because they enable parallel token decoding, but the large number of visual tokens still significantly hinders their inference efficiency. While visual token pruning has been extensively studied for autoregressive VLMs (AVLMs), it remains largely unexplored for DVLMs. In this work, we propose RedVTP, a response-driven visual token pruning strategy that leverages the inference dynamics of DVLMs. Our method estimates visual token importance using attention from the masked response tokens. Based on the observation that these importance scores remain consistent across steps, RedVTP prunes the less important visual tokens from the masked tokens after the first inference step, thereby maximizing inference efficiency. Experiments show that RedVTP improves token generation throughput of LLaDA-V and LaViDa by up to 186% and 28.05%, respectively, and reduces inference latency by up to 64.97% and 21.87%, without compromising-and in some cases improving-accuracy.
The growing concern about data privacy has led to the development of private inference (PI) frameworks in client-server applications which protects both data privacy and model IP. However, the cryptographic primitives required yield significant latency overhead which limits its wide-spread application. At the same time, changing environments demand the PI service to be robust against various naturally occurring and gradient-based perturbations. Despite several works focused on the development of latency-efficient models suitable for PI, the impact of these models on robustness has remained unexplored. Towards this goal, this paper presents RLNet, a class of robust linearized networks that can yield latency improvement via reduction of high-latency ReLU operations while improving the model performance on both clean and corrupted images. In particular, RLNet models provide a "triple win ticket" of improved classification accuracy on clean, naturally perturbed, and gradient-based perturbed images using a shared-mask shared-weight architecture with over an order of magnitude fewer ReLUs than baseline models. To demonstrate the efficacy of RLNet, we perform extensive experiments with ResNet and WRN model variants on CIFAR-10, CIFAR-100, and Tiny-ImageNet datasets. Our experimental evaluations show that RLNet can yield models with up to 11.14x fewer ReLUs, with accuracy close to the all-ReLU models, on clean, naturally perturbed, and gradient-based perturbed images. Compared with the SoTA non-robust linearized models at similar ReLU budgets, RLNet achieves an improvement in adversarial accuracy of up to ~47%, naturally perturbed accuracy up to ~16.4%, while improving clean image accuracy up to ~1.5%.