NEO Semiconductor to show 3D memory with AI processing
At FMS 2024 (Future of Memory and Storage), taking place in Santa Clara, California, NEO Semiconductor’s CEO, Andy Hsu, will introduce the 3D X-AI™, a game-changing 3D memory with AI processing that combines data storage and data processing in a single chip, accelerating neural performance by 100X and reducing power consumption by 99%.
Based on NEO’s award-winning 3D X-DRAM™ technology, 3D X-AI simulates artificial neural networks (ANNs), including synapses for weight data storage and neurons for data processing, making it ideally suited to accelerate next-generation AI chips and applications. 3D X-AI substitutes for high-bandwidth memory (HBM) and is expected to significantly evolve AI chip design and AI workload optimization.
A single 3D X-AI die includes 300 layers of 3D memory with 128 Gb (gigabit) capacity and 1 layer of neural circuitry with 8,000 neurons supporting up to 10 TB/s (terabytes per second) of AI processing throughput per die. Using twelve 3D X-AI dies stacked with HBM packaging can scale 3D X-AI chip capacity and performance by 12X to reach 1,536 Gb (192 GB) capacity and 120 TB/s processing throughput with a single 3D X-AI chip.
“Typical AI chips use processor-based neural networks. This involves combining high-bandwidth memory to simulate synapses for storing weight data and graphical processing units (GPUs) to simulate neurons for performing mathematical calculations. Performance is limited by the data transfer between HBM and GPU, with back-and-forth data transfer lowering AI chip performance and increasing power consumption,” said Andy Hsu, Founder & CEO of NEO Semiconductor. “AI chips with 3D X-AI use memory-based neural networks. These chips possess neural network functions with synapses and neurons in each 3D X-AI chip. They are used to drastically reduce the heavy workload of data transfer data between GPU and HBM when performing AI operations. Our 3D memory drastically improves AI chip performance and sustainability”.
If you enjoyed this article, you will like the following ones: don't miss them by subscribing to :
