Parallel Vector Processing in Optical Computing With Jacobi Time‐Wave Packets

ABSTRACT

The computational demands of artificial intelligence increasingly challenge the speed and energy efficiency of electronic processors, particularly for multiply‐accumulate (MAC) operations fundamental to neural networks. Current photonic accelerators face scalability and power consumption limitations. Here we demonstrate that Jacobi time‐wave packets (JTP), based on the superposition of n + 1 complex‐weighted orthogonal base functions, allow the simultaneous execution of 2( n + 1) MAC operations in a single clock cycle using a single modulator. Experimentally, we achieve the superposition of six weighted base functions for the real‐time unscrambling of 2 GBd 6×6 MIMO signals and the nonlinear optical vector processing for XOR and XNOR logic gates across four 10 GBd wavelength channels. This approach reduces processing steps and hardware complexity, offering a scalable and energy‐efficient pathway for optical computing in large‐scale neural networks. Our findings suggest that JTP can enhance the performance and integration density of photonic processors for AI applications and optical computing.