Inside the compute
Small array. Real arithmetic.
Follow a matrix multiplication through a grid of stationary weights.
Local simulation · no dependencies
Processing array
5 × 5 PEsA column vectors →Weights stay. Activations move.
CYCLE 0 / 9 compute cycles
ReadyReplay starts at zero. Step pauses playback and advances one clock tick.
Output matrix C
A × BComputed from the processing array.
Every cell is a sum of products. Negative values and zeros take the same path through the array.
C[i,j] = Σₖ A[i,k] × B[k,j]
Inspect output row
PE partial sums show the selected output row.
Inspect output row
PE partial sums show the selected output row.
The inputs
A Activations · integers −9…9
×
B Stationary weights
01Load the weights
Each PE keeps one B value. It stays put for the entire computation.
02Stream the activations
Row k receives A[:,k] at cycle k + 1. The vector moves one column right per tick.
03Accumulate the result
Each PE multiplies every vector lane by its weight. Column accumulators sum those contributions into C.