A parameterized multi-cycle Harvard-architecture processor built from 12 RTL modules, with an Amdahl's Law-backed study of a control-path optimization.
Six 64-bit adder topologies characterized on identical terms, from Ripple-Carry to Kogge-Stone, mapping the area/speed/power trade-off in carry computation.
Four 32-bit multiplier architectures compared by latency and throughput rather than clock speed alone, including the effect of the final adder stage.
A parameterized binary Hopfield neural network in SystemVerilog, taken from RTL through synthesis and physical implementation to GDSII, recovering stored patterns from noisy or incomplete inputs.
A binary classifier that corrects its own decision boundary from feedback in pure deterministic hardware, with two algorithms and formal convergence proofs.
Synchronous, asynchronous, and elastic FIFOs in Chisel, showing how the architecture changes with clocking, clock-domain crossing, and control flow.
A 4-bit feedback-driven machine that evolves across five versions, from a single self-correcting adder to a stored-program automaton executing instructions from memory.
A custom assembly language and Python assembler for the machine above, with lexical, syntax, and semantic analysis built up across three versions.
A gate-level McCulloch–Pitts neuron in Verilog, reproducing AND, OR, and NOT through threshold and inhibitory-weight adjustment alone.