| Field | Value |
|---|---|
| Application | Pulp |
| Technology | 130 |
| Manufacturer | IHP |
| Type | Teaching |
| Package | QFN56 |
| Dimensions | 2235 µm × 2235 µm |
| Gates | 350 kGE |
| Voltage | 1.2 V |
| Power | 47.4 mW @ 80 MHz |
| Clock | 80 MHz |
Modern embedded systems often require real-time processing of error-corrected sensor streams. Crocodilo accelerates a workload of this kind, built around three main stages:
- Decoding a signal composed of 64 samples of 32-bit words, protected with a Hamming (32, 26) Error-Correction Code (ECC).
- Computing an FFT (Fast Fourier Transform) on the decoded signal.
- Encoding the processed signal using the same Hamming ECC scheme.
A full software implementation of the workload was first developed to run on the original Croc design (with minor modifications), serving as a baseline. The hardware was then enhanced in four key ways:
- A custom memory-mapped ECC accelerator added to the user domain of the SoC.
- A fast multiplier integrated into the CVE2 core.
- SRAM capacity increased to 16 KB.
- Clock frequency raised to 94.3 MHz through timing optimizations.
The full description of the project can be found in the final report
Crocodilo achieves a 37× speed-up on the real-time signal-processing workload, reducing end-to-end latency from 5.2 ms to 142 μs. This gain is driven by:
- A single-cycle ECC accelerator, improving encode/decode throughput by over 40×.
- A fast multiplier unit, accelerating the FFT stage by nearly 30×.
- Expanded 16 KB SRAM, enabling the full memory footprint of the application.
- A higher operating frequency of 94.3 MHz.
This chip was designed as part of the VLSI design course at ETH Zurich which uses a (mostly) open source design flow for its exercises. Students are required to modify a Croc based SoC to improve its capabilities somehow to pass the course. This was one of the top-rated designs from the course and has been sent to manufacturing.
This work received generous support from the Leibniz Institute for High Performance Microelectronics through the BMBF project FMD-QNC (16ME0831).