An all-optical general-purpose CPU and optical computer architecture
Claims of an all‑optical, general‑purpose CPU have reignited debate over whether photonic computing can ever match electronic processors on density, efficiency, and practicality. Commenters highlight fundamental limits such as optical wavelength, weak photon–photon interactions, and difficult memory design, suggesting optics is more promising for interconnects or specialized accelerators than for full CPUs. The thread also raises concerns about private companies publishing unreviewed, paper‑styled preprints on arXiv, alongside broader skepticism about the state and value of traditional peer review.
Status of the Paper and arXiv Use
- Many comments question a private company putting what feels like a press release into an arXiv-style LaTeX paper, without peer review.
- Concerns that arXiv is being “polluted” by PR-like manuscripts, making it harder for non-experts and some investors to distinguish solid science from hype.
- Counterpoint: technically rich “PR papers” at least expose enough detail for real experts to spot nonsense, unlike traditional marketing blurbs.
- Several posts argue that formal peer review itself has serious credibility problems (citation extortion, gatekeeping), so its absence is not a reliable negative signal.
Fundamental Limits of Optical Computing
- Strong claim: general-purpose, large-scale integration is impossible (or extremely unlikely) because optical feature sizes are limited by wavelength (~1.5 µm), orders of magnitude larger than modern electronic transistors.
- This implies optical CPUs of equivalent complexity would be enormous.
- Others note potential mitigations: wavelength-division multiplexing (many channels per waveguide), vertical stacking, plasmonics, metamaterials, and near-field/superlens techniques.
- Disagreement over whether higher clock rates, lower heat, and richer interconnect can realistically offset low device density.
Switching, Nonlinearity, and Efficiency
- Some readers look for a clear explanation of the “optical transistor” / switching mechanism; one points to semiconductor optical amplifiers and cross-gain/phase modulation as the canonical approach.
- Another line of critique cites prior work arguing optical transistors are fundamentally inefficient: photons interact weakly and require strong nonlinearities (high intensities), so energy advantages may be overstated.
- Use of 2D materials is viewed skeptically as a hand-wavy fix with limited interaction volume.
Architecture, Memory, and SUBLEQ Demo
- The paper’s architecture is framed as relevant only after optical interconnects are already mainstream; current work is a conceptual “next phase” focused on energy efficiency.
- Actual hardware demo is a 2‑bit SUBLEQ one-instruction machine using optical components; critics note it cannot realistically “run Doom” and relies on slow delay-line memory and ROM-like or phase-change-style storage.
- Some appreciate the architectural exploration and especially a figure plotting compute vs power vs global energy limits, calling it a useful big-picture framing.
Overall Sentiment
- Mix of enthusiasm for the conceptual work and diagrams, and strong skepticism about practicality, scalability, and the risk of overhyping optical general-purpose CPUs.