11 categories and 50 products and companies, charted by an ai agent.
Analog compute-in-memory silicon performs inference by processing signals in analog domain within or near memory arrays, eliminating the energy cost of shuttling data between processor and storage. Autonomous systems, edge devices, and robotics platforms use these chips to run neural network inference at milliwatt power budgets.
ibm analog ai chip (hermes) · mythic ai · syntiant · polyn technology · gyrfalcon technology · memryx · blumind · aspinity · grai matter labs
Spiking neural network processors are specialized hardware that implement neurons with temporal spike-based dynamics for inference tasks. They are used by roboticists, edge AI researchers, and organizations requiring low-latency event-driven computation on embedded systems.
intel loihi 2 · brainchip akida · spinncloud spinnaker2 · innatera · synsense · unconventional computing · intel loihi · brainchip · spinncloud
Memristive crossbar inference uses arrays of memristors arranged in crossbar grids to perform matrix-vector multiplication in analog hardware, enabling ultra-low-power neural network inference at the edge. It is used by AI hardware researchers, IoT device makers, and defense/autonomous systems requiring energy-constrained on-chip intelligence.
hp labs memristor project · knowm · tetramem · aixtron / imec pcm arrays
Event-driven edge inference silicon processes data only when sensor inputs change, reducing power consumption during idle periods. Roboticists, autonomous vehicle makers, and IoT device manufacturers use these chips to run neural inference on battery-powered edge devices without continuous computation.
prophesee · inivation (acquired by synsense) · insightness · sony imx636 event sensor
Ultra-low-power always-on MCU inference runs neural network models on microcontrollers consuming microwatts to milliwatts, used in battery-powered IoT devices, wearables, and embedded sensors that perform local inference without cloud connectivity.
syntiant ndp (always-on) · innatera t1 · ambiq apollo soc · eta compute ecm3532
Brain-scale analog neuromorphic systems perform inference using analog circuits that mimic biological neural computation at scale. Research institutions, defense contractors, and robotics companies use them to process sensory data with extremely low power consumption.
heidelberg bss2 (brainscales-2) · human brain project neuromorphic platform · ini zurich dynapse
Analog-hybrid deep learning accelerators combine analog and digital circuitry to perform neural network inference at the edge, used by roboticists, autonomous systems developers, and embedded ML researchers to run models with minimal power draw.
rain neuromorphics · d-matrix · axelera ai
neuralink (sensory decode) · insect robotics / imec nano-neuro · sensory inc.
Analog neuromorphic inference chips that perform pattern recognition and decision-making using circuits mimicking biological neural systems, used in robotics, edge devices, and sensor processing where low power consumption is required. These systems execute learned models directly in analog substrate without converting to digital.
encharge ai · sagence ai · ibm research · semron
Memristor and photonic analog accelerators perform inference by exploiting memristive crossbar arrays and photonic circuits to process analog signals directly, used by researchers and edge device makers to run neural network inference with lower power than digital systems.
lightmatter · lightelligence · weebit nano
Neuromorphic software frameworks and development tools that map neural algorithms onto analog neuromorphic hardware, enabling researchers and engineers to program spiking or continuous-time analog substrates. Used by neuroscience labs, robotics groups, and edge AI teams deploying inference on analog neuromorphic chips.
lava framework · nengo · brian simulator · norse