A CPU that thinks in −1, 0, +1 — not just 0 and 1.
The 5500FP is a 24-trit RISC processor running on genuine ±3.3V ternary signal levels, not an emulation layer over binary logic. Built for supercomputing, AI and brain-simulation workloads that outgrow binary — the same architecture also lays the groundwork for edge and low-power applications.
Run entire operating systems, complex algorithms and computational models natively in a pure three-state environment — every layer of the stack handled directly in ternary, with no binary translation underneath.
Each trit carries roughly 1.58 bits of information, enabling compact data structures for finite automata, formal parsing and multi-valued routing — expressed in their natural logic paradigm, not forced into binary.
Ternary logic maps directly onto ternary neural networks and modern LLM weight quantization (−1, 0, +1). Executing these natively removes the binary-emulation bottleneck for symbolic AI and alternative ML models.
As transistors approach atomic scale, energy consumption and heat generation — not raw transistor count — have become the limiting factors. Ternary systems raise the information density per connector, so the same physical bus carries more state.
Ternary computing has mostly stayed an academic research topic since binary became the mainstream default in the 1970s. Building it forward means building the infrastructure and tooling almost from scratch — but it also means skipping decades of binary legacy constraints and designing directly for what's next.
A 24-trit balanced-ternary RISC processor — the hardware realization of our ternary architecture, with real ±3.3V ternary signal levels on the external buses.
Real hardware with a real ternary processor inside.
Open for preorder"5500FP: A 24-Trit Balanced Ternary RISC Processor," published on Zenodo — the full technical account of the architecture and its FPGA implementation.
Presented at the 35th International Workshop on Post-Binary ULSI Systems, held alongside the IEEE International Symposium on Multiple-Valued Logic.