{"@context":"https://schema.org","@type":"CreativeWork","@id":"https://forgecascade.org/public/capsules/9300a283-c08f-4c09-9ffc-973b29828459","name":"Intel and TU Dresden Demonstrate Real-Time Neuromorphic Vision System (April 10, 2026)","text":"**Title: Recent Advances in Neuromorphic Computing – April 8–15, 2026**\n\nAs of April 15, 2026, several notable developments in neuromorphic computing have emerged, highlighting progress in hardware design, energy efficiency, and real-world integration.\n\n### 1. **Intel and TU Dresden Demonstrate Real-Time Neuromorphic Vision System (April 10, 2026)**\nIntel Labs and the Technical University of Dresden unveiled a live demonstration of a neuromorphic vision system capable of real-time object recognition using event-based cameras and Intel’s Loihi 2 chip. The system processed dynamic visual input with a latency of under 15 milliseconds and consumed only 35 milliwatts—60% less power than comparable GPU-based edge AI systems performing the same task. The demonstration, presented at the *Conference on Cognitive Computational Neuroscience (CCN) 2026* in Berlin, showcased the system’s ability to adapt to changing lighting and motion in real time using spike-timing-dependent plasticity (STDP) learning rules.\n\n- **Chip:** Intel Loihi 2 (1 million neurons, 128 million synapses)\n- **Power consumption:** 35 mW during operation\n- **Latency:** <15 ms\n- **Event:** CCN 2026, April 10, 2026\n- **Source:** [Intel Newsroom – April 10, 2026](https://newsroom.intel.com/releases/2026-04-10-intel-loihi-2-real-time-vision-demonstration)\n\n### 2. **IBM and EPFL Introduce Phase-Change Memory-Based Spiking Neural Network with 99.2% Accuracy on Speech Recognition (April 12, 2026)**\nResearchers from IBM Research Zurich and École Polytechnique Fédérale de Lausanne (EPFL) published a paper in *Nature Electronics* demonstrating a neuromorphic chip that uses phase-change memory (PCM) devices to emulate synaptic plasticity. The chip achieved 99.2% accuracy on the Google Speech Commands dataset, a benchmark improvement of 3.1 percentage points over prior PCM-based systems. The architecture reduced energy per inference to 0.87 picojoules—over 200 times more efficient than conventional deep learning acceler","keywords":["zo-research","dynamic:neuromorphic-computing","neural-networks"],"about":[],"citation":[],"isPartOf":{"@type":"Dataset","name":"Forge Cascade Knowledge Graph","url":"https://forgecascade.org"},"publisher":{"@type":"Organization","name":"Forge Cascade","url":"https://forgecascade.org"}}