Opticore develops photonic chips designed to accelerate AI workloads by utilizing light-based signal processing instead of traditional electrons. These chips perform matrix operations with significantly higher speed and energy efficiency compared to conventional electronic processors. The technology is purpose-built to meet the high throughput and density demands of modern AI inference applications.
Funding
$5M raised to dateRaised to date based on public sources. This may differ from the amount the company actually raised and is based only on what is publicly available on the internet.
Founders
Product
Problem
The increasing demands of AI and machine learning workloads are pushing traditional CMOS-based processors to their limits due to the slowdown of Moore's Law and the breakdown of Dennard scaling. Data movement with electronic wires leads to capacitive loss, limiting clock speeds and causing significant thermal dissipation, creating bottlenecks in energy efficiency and computing density.
Solution
Opticore addresses these challenges with its patented optical processing units (OPUs) that utilize photonic integrated circuits for on-chip data movement and computation, specifically designed for memory-intensive AI tasks. By converting memory data to optical beams, the OPUs eliminate capacitive resistance, enabling faster clock speeds and significantly reduced energy consumption. This approach allows for potentially unlimited memory capacities and promises breakthroughs in high-performance computing for data centers by achieving 100x higher energy efficiency and computing density compared to digital electronics. The chips are fabricated using standard foundry services, with co-packaged optoelectronics and hyperbonding of high-bandwidth memories (HBMs) for volume production.
Target Audience
Opticore's primary customers are data centers and organizations involved in high-performance computing that require energy-efficient and high-density computing solutions for AI and machine learning applications.
Features
- Photonic logic based on photoelectric multiplication for low-energy computing
- On-chip data movement and computation with photonic integrated circuits
- Compatibility with high-bandwidth memories (HBMs) for scalability and performance
- Temporal mapping to encode neural data using optical pulses, enabling processing of trillions of parameters
- Dynamically programmable parameters for training tasks
- Fabrication using standard CMOS foundry processes with co-packaged optoelectronics