5 March 2024 in Journal of Lightwave Technology
With the continuous increase in the size and complexity of machine learning models, the need for specialized hardware to efficiently run such models is rapidly growing. To address such a need, silicon-photonic-based neural network (SP-NN) accelerators have recently emerged as a promising alternative to electronic accelerators due to their lower latency and higher energy efficiency. Not only can SP-NNs alleviate the fan-in and fan-out problem with linear algebra processors, their operational bandwidth can match that of the photodetection rate (typically $\approx$ 100 GHz), which is at least over an order of magnitude faster than electronic counterparts that are restricted to a clock rate of a few GHz. Unfortunately, the underlying silicon photonic devices in SP-NNs suffer from inherent optical losses and crosstalk noise originating from fabrication imperfections and undesired optical couplings, the impact of which accumulates as the network scales up. Consequently, the inferencing accuracy in an SP-NN can be affected by such inefficiencies—e.g., can drop to below 10%—the impact of which is yet to be fully studied. In this paper, we comprehensively model the optical loss and crosstalk noise using a bottom-up approach, from the device to the system level, in coherent SP-NNs built using Mach–Zehnder interferometer (MZI) devices. The proposed models can be applied to any SP-NN architecture with different configurations to analyze the effect of loss and crosstalk. Such an analysis is important where there are inferencing accuracy and scalability requirements to meet when designing an SP-NN. Using the proposed analytical framework, we show a catastrophic inferencing accuracy drop of up to 84% for SP-NNs of different scales with three known MZI mesh configurations (i.e., Reck, Clements, and Diamond) due to accumulated optical loss and crosstalk noise.
23 November 2022 in IEEE Transactions on Nanotechnology
Silicon-photonic neural networks (SPNNs) and artificial intelligence (AI) accelerators have emerged as promising successors to electronic accelerators by offering orders of magnitude lower latency and higher energy efficiency. Nevertheless, the underlying silicon photonic devices in SPNNs are sensitive to inevitable fabrication-process variations (FPVs) stemming from optical lithography imperfections. Consequently, the inferencing accuracy in an SPNN can be highly impacted by FPVs—e.g., can drop to below 10%—the impact of which is yet to be fully studied. In this paper, we, for the first time, model and explore the impact of optical phase noise due to FPVs in the waveguide width, silicon-on-insulator (SOI) thickness, and etch depth in coherent SPNNs that use Mach–Zehnder interferometers (MZIs). Leveraging such models, we propose a novel variation-aware, design-time optimization solution to improve MZI tolerance to different FPVs in SPNNs. Simulation results for two example SPNNs of different scales under realistic and correlated FPVs indicate that using the optimized MZIs can lead to significant improvements in the network inferencing accuracy. The proposed one-time optimization imposes low area overhead and hence is applicable even to resource-constrained designs.
2 August 2022 in Journal of Lightwave Technology
Integrated photonic neural networks (IPNNs) are emerging as promising successors to conventional electronic AI accelerators as they offer substantial improvements in computing speed and energy efficiency. In particular, coherent IPNNs use arrays of Mach–Zehnder interferometers (MZIs) for unitary transformations to perform energy-efficient matrix-vector multiplication. However, the underlying MZI devices in IPNNs are susceptible to uncertainties stemming from optical lithographic variations and thermal crosstalk and can experience imprecisions due to non-uniform MZI insertion loss and quantization errors due to low-precision encoding in the tuned phase angles. In this paper, we, for the first time, systematically characterize the impact of such uncertainties and imprecisions (together referred to as imperfections) in IPNNs using a bottom-up approach. We show that their impact on IPNN accuracy can vary widely based on the tuned parameters (e.g., phase angles) of the affected components, their physical location, and the nature and distribution of the imperfections. To improve reliability measures, we identify critical IPNN building blocks that, under imperfections, can lead to catastrophic degradation in the classification accuracy. We show that under multiple simultaneous imperfections, the IPNN inferencing accuracy can degrade by up to 46%, even when the imperfection parameters are restricted within a small range. Our results also indicate that the inferencing accuracy is sensitive to imperfections affecting the MZIs in the linear layers next to the input layer of the IPNN.
16 March 2020 in IEEE Transactions on Very Large Scale Integration (VLSI) Systems
Carbon-nanotube FETs (CNFETs) are potential successors to CMOS transistors; these emerging devices have a low intrinsic delay due to near-ballistic transport in carbon nanotubes (CNTs). As CNFETs are evaluated for circuit/system design, it is important to analyze variations in CNT process parameters. In this article, we present a systematic approach to quantify the impact of these imperfections on the transistor- and gate-level performances of CNT-based circuits. Process variations are investigated to identify the critical device parameters that have maximum impact on the device on-current. We also present a model that predicts the realistic CNFET yield in the presence of process variations. Finally, the impact of manufacturing defects, such as pinholes in the gate dielectric and parasitic CNFETs formed due to imperfect etching, are modeled and evaluated using HSPICE.
7 February 2023 in IEEE Journal of Selected Topics in Quantum Electronics
Coherent integrated photonic neural networks (IPNNs) are increasingly being explored for rapidly growing artificial intelligence applications. However, the principal roadblocks in the scalability of IPNNs are their large area footprint and high tuning power consumption during both training and inferencing. In deep neural networks (DNNs), software techniques to prune redundant weights are often utilized to reduce resource (e.g., memory, computation, and power) overheads. However, due to the complex nature in which the software weights are mapped to the building blocks of IPNNs, prior efforts to apply existing pruning approaches to IPNNs have been ineffective. We present CHAMP and LTPrune, two novel hardware-aware pruning techniques for IPNNs. Using a case study of three IPNNs with different footprints, we show that both these methods can prune more than 99% of the phase angles (which are similar to the weight parameters in DNNs). We also analyze the performance of the pruned IPNNs under phase uncertainties and present a comparative analysis of the two methods to enable advanced hardware-software-assisted design-optimization techniques for IPNNs. To expedite pruning, we also propose HybridPrune, where CHAMP and LTPrune are used in conjunction to obtain similar network sparsity as standalone-LTPrune but with up to 78.3% fewer retraining epochs.