Scientists Create Ultra Fast Memory Using Light

Light-powered memory arrives—AI giants drool, old pros roll eyes

TLDR: Scientists built a foundry-made memory bit that stores data as light, promising faster, cooler computing. The comments are split: fans say this could fix bandwidth woes for AI, skeptics argue optics are power-hungry and MRAM will win, with jokes about typos and OpenAI hoarding everything.

USC ISI and UW–Madison just dropped a “photonic latch” — a tiny memory cell that stores data as light and even refreshes itself — built on a standard silicon foundry, not sci‑fi lab gear. In plain speak: it’s a real light-powered memory bit, the missing piece for faster, cooler computing without constantly converting light to electricity. The team says it could scale into optical SRAM (the fast memory near your CPU), potentially turbocharging AI and data centers. Cue the internet battle royale.

The hype squad cheered: “Bandwidth is the bottleneck—finally, a fix!” Others slammed the brakes: veteran voices warned digital optical computing can be “immensely power hungry,” citing classic research, and one commenter insisted MRAM (magnetic RAM) and compute-in-memory are a decade ahead. Meanwhile, the clown car pulled up with jokes: a sharp-eyed reader mocked a “300mm chips” typo as proof “nobody’s checking the AI output,” another quipped that OpenAI will buy the world’s entire supply, and a third said the headline felt like a hallucinated future from a viral post. The spiciest mini-feud? Whether this is real hardware or just simulations — the article says foundry-made, skeptics say "we’ve seen flashy optics demos for decades." Buckle up, the light wars are on.

Key Points

  • USC ISI and UW–Madison developed a regenerative photonic memory (“photonic latch”) on a commercial silicon photonics platform.
  • The device stores data as light and regenerates signals to resist noise, analogous to electronic RAM but faster.
  • It aims to address interconnect delay and energy waste from optical–electrical domain switching in current systems.
  • Researchers published simulations showing how the latch could scale to a complete photonic SRAM system.
  • The work uses industry-standard processes and references GlobalFoundries for commercial scalability.

Hottest takes

"OpenAI has already committed to buying the entire world’s supply" — xienze
"MRAM-CIM is like 10 years ahead of this" — ilaksh
"Digital optical computing will be immensely power hungry" — cycomanic
Made with <3 by @siedrix and @shesho from CDMX. Powered by Forge&Hive.