UCLA researchers have developed cascadable, all-optical NAND gates built from passive diffractive neural networks, enabling logic operations (e.g. NAND, AND, OR, half-adder) via light propagation and diffraction rather than electronics, potentially lowering latency and power consumption.
Traditional optical logic gates often rely on nonlinear materials, interferometers, photonic crystals, or active components, which impose constraints on input signal intensity, require precise phase or polarization control, and reduce cascadability (i.e. difficulty in chaining gates without signal degradation). This limits their viability for scalable optical computing.
The team designed and numerically optimized a diffractive neural network composed of four passive layers. They encode logical inputs as optical power in two spatially separated apertures, train the network to perform NAND logic, and use the same NAND modules cascaded in sequence to build more complex logic (AND, OR) and arithmetic (half-adder) circuits. Importantly, they introduced a design map to avoid cascaded input/output optical field combinations that lead to inference errors, improving accuracy and enabling feasible optical logic chaining—without relying on nonlinear optics.
Passive optical components: no need for active/nonlinear materials for basic logical operations.
Cascadability: same NAND design reused to build larger logic functions.
Low latency and potential power efficiency compared to electronics.
Robust signal encoding (via relative power in apertures) tolerates non-ideal input uniformity.
Scalability of logic circuits (e.g. half-adder) built from simple modular units.
Design guidance (error maps) that helps avoid input configurations causing logic errors.
Low-power optical computing platforms where traditional electronics are power- or latency-limited.
Edge computing, photonic processors, or integrated optical logic circuits.
Optical signal processors in imaging, communications, or sensor networks.
Components in all-optical neural network or deep learning hardware.
Secure or radiation-resistant computing where optical paths are more robust.
Luo, Y.; Mengu, D.; Ozcan, A. (2022). Cascadable all-optical NAND gates using diffractive networks. Scientific Reports, 12, Article 7121. DOI: 10.1038/s41598-022-11331-4