Using light instead of electrons is an obvious, ideal form of signaling that promises more bandwidth and more power efficiency. However, the complexities of using such a delicate form of energy have meant that the only present deployment of photonics is in optical cables, from household ethernet to those spanning the distance between continents.
While those are cool, it’s the board-scale and chip-scale uses that have made photonics such a hot topic. In short, light emissions have low power thresholds so signals are sent earlier and faster, resulting in on-chip latencies orders of magnitude better than electrical signals. The issue is that right now, converting an optical signal to an electrical one requires a lot of power and space, negating all the gains.
We’ve seen Interesting attempts to remedy this issue in the past, but the new research tackles the problem in an entirely different way. If the memory (be it primary storage, ram or cache) can accept and output information in both forms, then there’s no need to convert it.
The scientists’ memory cell, which is a non-volatile germanium-based compound, sits at the intersection of gold electrodes and silicon-nitride channels. Electrons flow through the gold, and light waves are funneled through the channels. When either hits the cell, the cell switches between a binary or multi-level state. Your standard 0s and 1s stuff.
While the specific uses of this technology will hopefully materialize in the coming decades, scientists have envisioned how it could help solve current problems. The memory cells might facilitate the development of photonic transistors and could act as a cache and interface for reconfigurable photonic circuits and photonic neural networks. The sky’s the limit.