Siphi provides silicon photonics chips that combine photonic and electronic circuits to replace power‑hungry electronic interconnects in AI data‑center accelerators. Their turnkey modules include packaging, reference designs, and software libraries, delivering high‑bandwidth, low‑latency optical links that improve power efficiency and scalability for AI hardware vendors.
Funding
Funding not disclosed
Founders
Product
Problem
Current AI data center accelerators rely on electronic interconnects that consume significant power and face bandwidth limitations, restricting the scalability and efficiency of high‑speed AI workloads.
Solution
SiPhi creates silicon photonics chips that integrate photonic circuits with dedicated companion electronics, enabling optical data transmission at light speed for AI processing. By employing heterogeneous integration and a deep IP stack, the company delivers turnkey modules that include packaging, reference designs, and optimized software algorithms. These solutions replace traditional electronic interconnects with optical engines that offer higher bandwidth, lower latency, and reduced power consumption. The integrated approach simplifies deployment in data centers and emerging AI hardware, providing a scalable pathway to accelerate AI inference and training workloads.
Target Audience
Primary customers are AI accelerator manufacturers, data center hardware providers, and OEMs developing high‑performance AI compute platforms that require efficient optical interconnect solutions.
Features
- Heterogeneous integration of Photonic Integrated Circuits (PIC) and Electronics IC (EIC) on a single silicon substrate
- Deep IP stack covering photonic design, electronic control, and system-level interfaces
- Turnkey reference designs and packaging that accelerate time‑to‑market for AI hardware vendors
- Optimized software and algorithm libraries that map AI workloads onto the optical engine
- High‑bandwidth, low‑latency optical interconnects delivering multiple‑times improvement in power efficiency over electronic alternatives
- Scalable architecture designed for data center AI accelerators and emerging physical AI devices