Neuro-Photonix: Enabling Near-Sensor Neuro-Symbolic AI Computing on Silicon Photonics Substrate

TL;DR

Neuro-Photonix tackles the challenge of near-sensor neuro-symbolic AI for IoT by leveraging a silicon-photonic substrate to perform neural dynamics and HyperDimensional Computing-based symbolic reasoning with minimal ADC/DAC reliance. The architecture combines a Low-overhead Modulation Unit and Reconfigurable Optical Core to execute MAC operations in a single photonic cycle, while HD encoding maps neural outputs into robust HVs, enabling edge reasoning and reduced cloud dependency. Through a bottom-up evaluation framework, the work demonstrates strong energy efficiency (up to $30$ GOPS/W) and substantial reductions in data transmission and processing latency compared with both electronic and optical baselines, while preserving accuracy on reasoning tasks like RAVEN. The results indicate significant practical impact for edge AI in IoT, offering scalable near-sensor inference with interpretable symbolic reasoning and potential for further hardware-embedded HD similarity and approximate computing features.

Abstract

Neuro-symbolic Artificial Intelligence (AI) models, blending neural networks with symbolic AI, have facilitated transparent reasoning and context understanding without the need for explicit rule-based programming. However, implementing such models in the Internet of Things (IoT) sensor nodes presents hurdles due to computational constraints and intricacies. In this work, for the first time, we propose a near-sensor neuro-symbolic AI computing accelerator named Neuro-Photonix for vision applications. Neuro-photonix processes neural dynamic computations on analog data while inherently supporting granularity-controllable convolution operations through the efficient use of photonic devices. Additionally, the creation of an innovative, low-cost ADC that works seamlessly with photonic technology removes the necessity for costly ADCs. Moreover, Neuro-Photonix facilitates the generation of HyperDimensional (HD) vectors for HD-based symbolic AI computing. This approach allows the proposed design to substantially diminish the energy consumption and latency of conversion, transmission, and processing within the established cloud-centric architecture and recently designed accelerators. Our device-to-architecture results show that Neuro-Photonix achieves 30 GOPS/W and reduces power consumption by a factor of 20.8 and 4.1 on average on neural dynamics compared to ASIC baselines and photonic accelerators while preserving accuracy.

Figures (15)

• Figure 1: MR input and through ports’ spectra after imprinting a parameter (using tuning signal). By adjusting the MR's resonant wavelength ($\lambda_{res}$) using the phase shifter, part of the input signal drops into the ring (through the coupling region) towards the drop port while the remaining propagates towards the through port.
• Figure 2: Neuro-symbolic AI framework consisting of neural dynamic and symbolic AI models. Neural network models are responsible for information embedding, and symbolic models build up knowledge for symbolic decision-making.
• Figure 3: High-level operational flow of the proposed Neuro-Photonix architecture.
• Figure 4: Neuro-Photonix architecture consisting of a sensor array and the optical core.
• Figure 5: Schematic of (a) Comparator-based Convertor (CBC) and (b) Light Driver Unit (LDU), (c) Sample waveforms of CBC input from the pixel and respective outputs.