An AI neural network that assigns three values for its weights and activations between nodes (-1, 0, +1). A ternary neural network (TNN) lowers the required computations and memory usage compared to a full-precision model. See
AI weights and biases.
Ternary hardware that supports ternary computations is a promising new area, especially using optical components. Such hardware is implemented in custom-built ASIC chips as well as field-programmable FPGA chips. See
ternary chip,
ASIC and
FPGA.