Mushroom Memristors (semiengineering.com)

Researchers at Ohio State demonstrated that common edible mushrooms (shiitake and button) can act as organic memristors: they cultured and dehydrated fruiting bodies, attached electrodes to different anatomical points, and drove them with varying voltages and frequencies to elicit memristive switching. Configured as volatile RAM, fungal devices transitioned between electrical states up to 5.85 kHz with ~90% accuracy; performance degraded at higher speeds but improved by paralleling more mushroom elements. The team highlights shiitake’s radiation resistance for potential aerospace uses but notes major engineering challenges remain—device miniaturization, controlled cultivation and reproducibility—before fungal memristors could be practical biohybrid components. Two complementary hardware advances target power and compute bottlenecks in AI systems. UNIST researchers created an ultra-small hybrid LDO using a seamless digital-to-analog transfer (D2A-TF) and a local ground generator, achieving 0.032 mm² in 28 nm, 54 mV ripple during 99 mA swings, 667 ns transient recovery, and −53.7 dB PSRR at 10 kHz—aimed at transient stabilization for AI chips and 6G modules. Separately, a photonic convolution chip from UF/UCLA/GWU uses on-chip lasers and dual sets of miniature Fresnel lenses (fabricated with standard processes) to perform convolutions optically, attaining ~98% accuracy on handwritten-digit classification and offering wavelength-division parallelism for near-energy-free, high-throughput convolution acceleration. Together, these works point to diverse directions—biohybrid memory, compact power management, and photonic compute—for squeezing efficiency and density in future AI hardware.