SciSci Research
Website unreachable since Nov 2, 2026
SciSci is an AI chip design company that replaces traditional floating-point and fixed-point arithmetic with an error-free, p-adic based number system. This novel architecture enables exact arithmetic without rounding errors, delivering 15–50× faster compute speeds and up to 50× better energy efficiency for matrix multiplication units and tensor cores. The technology is available for licensing to AI accelerator manufacturers.
- Semiconductor
- Artificial Intelligence
Funding
Founders
Founder details are not available yet.
Product
Problem
Scaling AI compute is constrained by the inefficiencies and limitations of traditional floating-point and fixed-point arithmetic, which introduce approximation and rounding errors. These errors degrade performance and energy efficiency, limiting the speed and scale at which AI workloads can be processed.
Solution
SciSci Research is an AI chip design company with a novel architecture for radically faster and stronger arithmetic. The architecture replaces floating-point and fixed-point arithmetic with efficient, errorless compute based on p-adic analysis, a branch of modern number theory. This new arithmetic logic design and number representation format enables exact arithmetic without approximation or rounding error. By eliminating the inefficiencies of decimal-based approximation, SciSci's architecture is expected to deliver speeds 15–50× faster than floating-point on average, along with a 50× energy efficiency gain. The technology can be licensed for integration into matrix multiplication units, tensor cores, and other units used in AI chips.
Target Audience
Primary customers are AI chip manufacturers and accelerator designers seeking to scale AI workloads more quickly and cost-effectively through licensed arithmetic technology.
Features
- Replaces floating-point and fixed-point arithmetic with an error-free, p-adic based number system
- Enables exact arithmetic without approximation or rounding error
- Delivers 15–50× faster compute speeds than floating-point on average
- Provides up to 50× energy efficiency gain
- Can be licensed for use in matrix multiplication units, tensor cores, and other AI chip components