Sub-1-us, Sub-20-nJ Pattern Classification in a Mixed-Signal Circuit Based on Embedded 180-nm Floating-Gate Memory Cell Arrays
We have designed, fabricated, and successfully tested a prototype mixed-signal, 28x28-binary-input, 10-output, 3-layer neuromorphic network ("MLP perceptron"). It is based on embedded nonvolatile floating-gate cell arrays redesigned from a commercial 180-nm NOR flash memory. The arrays allow precise (~1%) individual tuning of all memory cells, having long-term analog-level retention and low noise. Each array performs a very fast and energy-efficient analog vector-by-matrix multiplication, which is the bottleneck for signal propagation in most neuromorphic networks. All functional components of the prototype circuit, including 2 synaptic arrays with 101,780 floating-gate synaptic cells, 74 analog neurons, and the peripheral circuitry for weight adjustment and I/O operations, have a total area below 1 mm^2. Its testing on the common MNIST benchmark set (at this stage, with a relatively low weight import precision) has shown a classification fidelity of 94.65%, close to the 96.2% obtained in simulation. The classification of one pattern takes less than 1 us time and ~20 nJ energy - both numbers much better than for digital implementations of the same task. Estimates show that this performance may be further improved using a better neuron design and a more advanced memory technology, leading to a >10^2 advantage in speed and a >10^4 advantage in energy efficiency over the state-of-the-art purely digital (GPU and custom) circuits, at classification of large, complex patterns.
