Abstract
Residue number system (RNS) enables dimensionality reduction of an arithmetic problem by representing a large number as a set of smaller integers, where the number is decomposed by prime number factorization using the moduli as basic functions. These reduced problem sets can then be processed independently and in parallel, thus improving computational efficiency and speed. Here we show an optical RNS hardware representation based on integrated nanophotonics. The digit-wise shifting in RNS arithmetic is expressed as spatial routing of an optical signal in 2x2 hybrid photonic-plasmonic switches. Here the residue is represented by spatially shifting the input waveguides relative to the routers outputs, where the moduli are represented by the number of waveguides. By cascading the photonic 2x2 switches, we design a photonic RNS adder and a multiplier forming an all-to-all sparse directional network. The advantage of this photonic arithmetic processor is the short (10's ps) computational execution time given by the optical propagation delay through the integrated nanophotonic router. Furthermore, we show how photonic processing in-the-network leverages the natural parallelism of optics such as wavelength-division-multiplexing or optical angular momentum in this RNS processor. A key application for photonic RNS is the functional analysis convolution with widespread usage in numerical linear algebra, computer vision, language- image- and signal processing, and neural networks.
Introduction
Traditional electronic processors are increasingly constrained by physical bottlenecks, notably resistance-capacitance (RC) delays and high power consumption during data movement. As modern artificial intelligence and signal processing tasks demand ever-greater computational throughput, researchers are turning to optical computing as a high-speed, low-energy alternative. However, standard binary arithmetic is notoriously difficult to implement efficiently in the optical domain due to the bulkiness and energy requirements of traditional components.
This paper presents a novel solution by combining the mathematical framework of the Residue Number System (RNS) with ultra-fast hybrid photonic-plasmonic (HPP) switches. By representing numbers as spatial positions of light and leveraging the inherent parallelism of optics, this architecture achieves picosecond-scale computation with unprecedented efficiency.
Architecture and Methodology
The proposed system relies on three core innovations:
- Residue Number System (RNS): RNS decomposes large integers into sets of smaller remainders (residues) based on prime moduli. Unlike binary systems, RNS operations are carry-free, allowing for massive parallelization as each residue can be processed independently.
- Hybrid Photonic-Plasmonic (HPP) Switches: The hardware utilizes 2×2 switches composed of silicon waveguides with an Indium Tin Oxide (ITO) switching island. This hybrid approach combines the low propagation loss of photonics with the high-speed, compact footprint of plasmonics, enabling femtojoule-level switching energy.
- All-to-All Sparse Directional (ASD) Schematic: To minimize component count, the authors developed the ASD schematic. This non-blocking router design significantly reduces the number of switches required compared to traditional mesh-grid layouts. For instance, a modulo-5 system requires only 10 switches instead of 20.
In this architecture, the system employs one-hot encoding, where a specific value is represented by the presence of light in a particular waveguide. Arithmetic operations—addition and multiplication—are then executed as simple "spatial routing" tasks. The computation speed is fundamentally limited only by the "time-of-flight" of light through the device, typically in the range of tens of picoseconds.
Performance and Key Results
The effectiveness of the ASD RNS model was validated through Finite-Difference Time-Domain (FDTD) simulations and mathematical modeling. The primary metric for comparison was the Speed-Energy-Area Product (SEAP) Figure of Merit (FOM), which accounts for the trade-offs between computational speed, energy consumption, and physical footprint.
Key findings include:
- 100x Efficiency Gain: The HPPS-based ASD model demonstrates a SEAP FOM roughly 100 times better than designs based on Micro-Ring Resonators (MRR) or Mach-Zehnder Interferometers (MZI).
- Ultra-Fast Response: The HPP switches offer a response time of approximately 5.1 ps, with a per-device propagation time of just 0.1 ps. This enables switching speeds up to 200 GHz.
- WDM Capability: Because the switches are broadband, the system can utilize Wavelength Division Multiplexing (WDM). Multiple "colors" of light can traverse the same hardware simultaneously, performing different calculations in parallel without interference.
Conclusion and Future Outlook
While the optical path is exceptionally fast, the current design requires an electrical control circuit and Look-Up Tables (LUTs) to manage switch states, which adds complexity as moduli scale. Future work will focus on integrating these components into full-scale processors capable of performing complex convolution operations. Such a system would be ideally suited for the high-speed Multiply-Accumulate (MAC) operations required in modern neural networks and real-time signal processing.
Summary updated