Neuromorphic computing builds hardware that computes the way brains are thought to: with spiking, event-driven, analog dynamics and weights stored where the arithmetic happens, and the demonstrated systems today range from a measured 100 W wafer-scale digital spiking machine to hundreds of small emerging-device arrays whose headline efficiency figures are extrapolations rather than measurements.
Summary
Evidence base
Open questions
Assessment drafted 2026-08-31 from up to 18 KB sources using the technology-scorecard framework; scores are a draft read pending review.
Recent mentions
2026-09-15A Unified Interconnection Network for Chiplet-Based Scaling of the BrainScaleS Neuromorphic Systempaper
2026-09-07Annual industrial robot shipments will reach more than 760k units in 2030 (Interact Analysis)web
2026-09-07CES 2026: Qualcomm targets NVIDIA Jetson with new robotics developer platform (A3 / automate.org)web
2026-09-07World Robotics 2025 report, service robots (IFR, 7 Oct 2025)web
2026-09-07Auterion secures contract to deliver 33,000 Skynode drone strike kits to Ukraine (Auterion; DroneLife, 29 Jul 2025)web
2026-09-07Nvidia's 'robot brain' chip Thor goes on sale (CNBC, 25 Aug 2025; NVIDIA newsroom)web
2026-09-07Microchip Technology signs definitive agreement to acquire Hailo (Microchip IR, 24 Jul 2026; CTech; Globes)web
Frontier questions
Does any emerging-device crossbar reach 10^4 to 10^6 cells with the device-to-device variability of the Ge3.5Te1 CBRAM result preserved, or does the variability advantage disappear with array size 2026 06 05 electrolyte bonding engineering for highly uniform gete base?
Can the 95% charge-recovery assumption underpinning the memcapacitor efficiency figures be measured on a real scaled device, given it alone accounts for a 17.4 times multiplier demasius 2021 memcapacitor nature electronics?
Is there a memory organisation that removes the new on-chip memory wall in digital neuromorphic processors, and what is the measured area and energy split after it is applied 2026 06 05 memory wall is not gone a critical outlook on memory archite?
Does any neuromorphic system, digital or analog, beat a commercial accelerator on a workload a buyer already runs, rather than on a task chosen to suit the hardware?
ANSWERED (20 Jul 2026, 2026 07 20 neuromorphic market sockets and algorithm vs hardware): Does any SPIKING chip find a volume socket beyond research/niche sensing? NO. The always-on edge volume sockets are real but won by conventional low-power digital (Ambiq 280M+, Syntiant 20M+) and sparse-digital (femtoAI 100k+), not spikes. Only defensible spiking wedge = defense RF signal classification, still POC (BrainChip FY25 revenue $1.89M). Base rate confirmed poor.
ANSWERED (20 Jul 2026): brain-inspired ALGORITHM (Camp A, >$2B in 2026 — Flourish, AMI/LeCun, Liquid, Sakana) runs on GPUs and does NOT need neuromorphic silicon; spiking HARDWARE (Camp B) is a separate ~$0.5B chip market. Never underwrite a hardware bet on the brain-inspired tailwind — it is a software story.
Do analog / memcapacitor devices become the substrate for neuromorphic, or stay separate?