Architect Labs Designs Redwood Chip in Two Weeks

This could be a major shift toward software-speed chip design, but the headline performance claims remain unproven until Redwood reaches manufacturing.

• –AI generated the RTL, verification environments, firmware, drivers, and compute kernels from a human-authored architecture.
• –Redwood reportedly runs Llama and Qwen models on AMD Versal FPGA hardware.
• –Claimed Samsung 8nm gains—1.75× throughput and 1.9× lower power than Jetson Orin Nano—are projections calibrated from FPGA measurements.
• –A model running on Redwood reportedly identified timing and kernel improvements, suggesting a feedback loop between AI workloads and hardware design.
• –Successful tapeout, yield, thermal behavior, and independent benchmarks will determine whether this is a breakthrough or an impressive prototype.