06 Oct 2026
Where the starter design’s 8,994 nJ goes
Trace per-layer area, power, time, and energy so the score stops being a mysterious number.
HLS Hackathon series
08 Oct 2026
Pruning in Python before touching HLS
Use parameter counts, multiplication counts, and repeated training runs to eliminate expensive designs before full hardware synthesis.
10 Oct 2026
How few bits can a small CNN survive on?
Walk through fixed-point width, fractional bits, overflow, saturation, rounding, and the accuracy tradeoff.
12 Oct 2026
Reuse factor: silent snapping and why 1 wins
Measure the area-versus-time tradeoff instead of assuming that fewer multipliers automatically means lower energy.
14 Oct 2026
RISC-V integration and submission
Follow the final integration step, record where the design landed, and turn the experiment into a reproducible result.
After 15 Oct
Full parameters, final rank, lessons
Publish the complete winning configuration and the lessons from the measured sweep after the contest closes.
What this section is for
These are engineering notes rather than marketing copy. The goal is to explain what changed, what was measured, what surprised me, and which mental model survived contact with the tools. The HLS work here follows the hackathon’s fixed flow: describe the network, train it, convert it to fixed point, synthesize with HLS, measure time and power, and reason from energy as power × time.