Matrix MAC Processing Element
Create the processing element used repeatedly in the matrix accelerator.
Task
Accumulate signed 16x16 products into a signed 40-bit accumulator with synchronous clear and enable.
Examples
- clear=1 starts a new dot product.
- enable=1 adds one product.
Constraints
- Signed arithmetic.
- 40-bit accumulation.
- Synchronous clear.
Accelerator verification checklist
- Exercise zero, maximum positive and negative operands where signed data is used.
- Check exact bit growth and make overflow or saturation behavior intentional.
- Verify reset, enable, valid and pipeline alignment cycle by cycle.
- Send back-to-back data to confirm the advertised throughput.
Concepts to review
matrix multiply, MAC, PE
Open Matrix MAC Processing Element in the HDLForge interactive editor.
Related problems
- Signed INT8 Multiply-Accumulate — Easy
- Four-Lane INT8 Dot Product — Medium
- Eight-Lane Reduction Adder — Medium
- ReLU with 8-bit Saturation — Medium
- Systolic Array Processing Element — Hard