All-optical AND Logic Gate Based on Semiconductor Optical Amplifiers for Implementing Deep Recurrent Neural Networks

As artificial intelligence models become increasingly complex, the hardware supporting them faces growing demands in terms of speed, energy efficiency, and parallel processing. Recurrent neural networks such as Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) architectures rely on internal gating mechanisms that repeatedly perform logical operations while processing sequential information. Although these operations are relatively simple, executing them electronically introduces latency and power consumption that can become significant in large-scale systems. The work presented here explores a radically different approach: implementing one of the fundamental logic operations required by recurrent neural networks entirely in the optical domain. Instead of converting optical signals into electrical ones for processing, the proposed architecture performs logical computation directly with light, opening a possible path toward future photonic neural computing systems.

The proposed device implements an all-optical AND logic gate by combining two well-established photonic concepts: an optical XOR stage and a pair of semiconductor optical amplifiers operating in a cross-phase modulation (XPM) configuration. The architecture begins by splitting two incoming optical signals into multiple branches. One branch generates an intermediate XOR output through constructive and destructive optical interference obtained by introducing a phase shift of pi-greco radians. This intermediate signal is then combined with the original inputs and injected into two parallel semiconductor optical amplifiers. Inside the amplifiers, the optical signals modify the refractive index of the semiconductor material through cross-phase modulation. As a consequence, the phase accumulated by each optical wave depends directly on the signal intensity. By carefully controlling these phase shifts, the optical outputs interfere again, producing the logical AND operation without requiring any electronic conversion.

Rather than analysing an idealised system, the proposed methodology incorporates a realistic numerical model of commercially available semiconductor optical amplifiers. The simulations include practical effects that strongly influence real optical communication systems, such as phase noise, white noise, gain saturation, signal distortion, and propagation delay. Device parameters are derived from published experimental data and commercial specifications, allowing the numerical model to closely reproduce the behaviour of physical optical amplifiers. This realistic modelling makes it possible to evaluate not only whether the logical function is theoretically achievable, but also how it would perform under operating conditions similar to those encountered in practical photonic hardware.

The proposed logic gate is analysed by testing all four possible combinations of two binary optical inputs. The simulations show that three of the four logical configurations produce nearly ideal behaviour. The ‘00’, ‘01’, and ‘11’ states generate the expected optical outputs with excellent extinction ratios and clean transitions. The remaining ‘10’ configuration is more challenging because the optical interference depends on a delicate balance between signal amplitudes and phase shifts inside the semiconductor amplifiers. To address this issue, the study investigates the relationship between amplifier saturation, optical gain, and induced phase variation, identifying the operating conditions required to minimise residual output power. The simulation results demonstrate that even this critical configuration achieves a significant extinction ratio, suggesting that further optimisation of amplifier design could lead to fully ideal behaviour.

The significance of this work extends beyond the implementation of a single logic gate. Logical operations such as AND constitute essential building blocks for approximating the gating mechanisms used inside recurrent neural networks, particularly when these models are translated into hardware operating with finite numerical precision. Demonstrating that such operations can be executed entirely with optical components suggests the possibility of constructing future photonic processors capable of performing neural computations at the speed of light. Unlike conventional electronic accelerators, these systems could exploit the intrinsic parallelism and bandwidth of optical signals while reducing latency associated with electrical conversions. Although further advances in semiconductor optical amplifier design will be required before complete all-optical recurrent neural networks become practical, the proposed architecture provides an important proof of concept. It shows that optical logic is not merely a theoretical possibility but a feasible technological direction for next-generation artificial intelligence hardware, where computation, communication, and signal processing converge within the same photonic platform.