This company designs dedicated Machine Learning hardware engineered as a pure, end-to-end AI accelerator. Their goal is to make running large-scale neural networks affordable, fast, and private by eliminating runtime software requirements. They focus on creating efficient compute solutions that support personal freedom and open access to AI technology.
Funding
Funding not disclosed
Founders
Product
Problem
Current AI development is constrained by the reliance on multi-layer software and expensive hardware for training and deploying models. This creates a significant barrier to entry and limits the efficiency, cost-effectiveness, and accessibility of advanced AI applications.
Solution
ThinkBe has developed a Hard-Core AI (HC-AI) chip featuring a novel Reconfigurable Neural Unit (RNU) circuit architecture. This design challenges the traditional balance between processing velocity, energy efficiency, and cost by hardcoding AI models into programmable circuit structures. By converting AI models directly into fixed circuit configurations, the HC-AI chip bypasses the need for extensive software layers and general-purpose CPUs/GPUs for inference. This approach enables local computation, enhancing data security and control, and significantly reduces energy consumption and cost compared to conventional solutions.
Target Audience
The primary target audience includes developers and organizations requiring high-performance, energy-efficient, and cost-effective AI processing for applications in machine learning, data analytics, and consumer electronics.
Features
- Programmable Hardcoded AI Chip with Reconfigurable Neural Unit (RNU) architecture.
- Achieves up to 97% energy reduction through gated circuits that model programmable neural network workflows.
- Reduces cost by an estimated factor of 20 compared to traditional AI hardware.
- Facilitates local computation, enhancing AI safety and data privacy by eliminating reliance on cloud services.
- Designed for scalability, with validation on FPGA micro-processors and planned transition to ASIC for enhanced performance.
- Enables direct input processing and fixed-link data transmission for high-speed inference.