Haskellator is an experimental tool designed to unlock new possibilities in processing RTL (Register-Transfer Level) netlists. It aims to enhance simulation efficiency, offer insightful change analysis, and explore experimental synthesis techniques.
### Key Features:
- **Sparse Simulation**:
Traditional open-source RTL simulators store signal values every time a signal changes.
HaskellatorSim will save signal states only if they’ve changed
within the last host CPU second of simulation. This approach compresses the simulation
data whilst enabling quick recomputation of any gaps when viewing saved simulations.
This requires access to the RTLIL sources and simulation save files to fill in any gaps.
- **Change Condition Analysis**:
Haskellator will provide tools to analyze and trace the cause of signal changes during
simulation. By forming hypotheses about which signals influenced a change, it offers a
"git-blame" style feature for simulations. This functionality also requires access to
the original RTLIL sources and simulation save files.
- **Experimental Synthesis Techniques**:
We hope to explore the potential of using reinforcement learning neural networks to
optimize synthesis passes for input netlists. The synthesized netlist could be placed
and routed to determine fmax, which could be used as the score the neural network will
learn to optimize.
These are just a few of the concepts we're experimenting with. The broader goal is to
explore innovative ideas, develop a high-quality tool that can evolve with community