Japan Is Building a 1.4nm AI Chip
Fujitsu is developing a neural processing unit (NPU) optimized for AI inference with transistors as small as 1.4 nanometers, enabled by the ambitious Rapidus semiconductor venture in Japan.
Why it matters
This breakthrough in transistor scaling could have a significant impact on the global chip supply chain and enable more efficient AI inference at the edge and in data centers.
Key Points
- 1Fujitsu is building a 1.4nm AI chip, smaller than a DNA double helix
- 2Rapidus, a new semiconductor company in Japan, is aiming to jump directly to 2nm chip manufacturing
- 3The 1.4nm chip is targeted for production around 2029, offering improved power efficiency for AI inference
Details
Fujitsu is developing a neural processing unit (NPU) optimized for AI inference with transistors as small as 1.4 nanometers, enabled by the ambitious Rapidus semiconductor venture in Japan. Rapidus has secured $1.7 billion in funding from the Japanese government and major companies to jump directly to 2nm chip manufacturing, skipping several intermediate nodes. The 1.4nm chip from Fujitsu is planned for production around 2029. This level of transistor density is critical for improving power efficiency and throughput for AI inference workloads, which prioritize low latency and high tokens-per-watt over the raw FLOPS needed for training large models.
No comments yet
Be the first to comment