Finchetto develops fully-optical passive network switches and photonic processors that significantly reduce energy consumption in data centers, which currently account for 2% of global electricity use. By achieving up to 26 times lower power requirements and 40 times reduced latency compared to traditional electronic switches, Finchetto enhances computational efficiency for AI applications.
Funding
Funding not disclosed
IIFounders
Product
Problem
Data centers, which are critical for AI and other computationally intensive applications, consume a significant and growing amount of global electricity, with network switches alone accounting for a substantial portion of this energy usage. Traditional electro-optical network switches present trade-offs between power consumption and latency, limiting overall computational efficiency.
Solution
Finchetto is developing fully-optical passive network switches and photonic processors that drastically reduce energy consumption and latency in data centers. By eliminating electronics in the switching process, Finchetto's technology achieves ultra-low latency and high throughput packet switching while requiring a fraction of the power compared to state-of-the-art network switches. This innovative approach enables much more compute for far less power, addressing the increasing demands of AI and other data-intensive applications. The fully optical, passive architecture is optimized for efficient AI, offering a new era of networking.
Target Audience
Finchetto's primary customers are data centers and hyperscalers seeking to improve energy efficiency and performance for AI and other computationally intensive workloads.
Features
- Fully optical, zero-electronics switching with completely passive logic
- Ultra-low port-to-port latency (40 nanoseconds), up to 50x lower than current state-of-the-art options
- Extreme efficiency, delivering a 26
- 53x reduction in power compared to state-of-the-art network switches (38 Watts)
- Unlimited bandwidth, enabling packet switching at the speed and throughput of the surrounding transceivers
- Compatibility with next-generation network topologies such as PolarFly for massively distributed compute