Architect Labs: an AI-designed chip in two weeks

Architect Labs claims to have designed and verified a chip named Redwood in two weeks, with the help of AI tools and two human architects. The design has not yet been manufactured, and the production planned at TSMC has no timeline. Tested on FPGA, Redwood could, according to the company, surpass the Jetson Orin Nano in efficiency and run models like Llama and Qwen. Academics find the speed impressive but reserve their judgment until manufacturing.

Zheng Makes Manufacturing the Decisive Test; TSMC Targeted Without Deadline

For Hao Zheng, an assistant professor at the University of Central Florida, manufacturing is the true test, revealing the speed, efficiency, and reliability of the chip. Redwood has not yet been manufactured. Ebrahim Hussain indicates that the company plans to send the design to TSMC, without a set date, as AI continues to improve the schematic. Matthew Guthaus at UC Santa Cruz emphasizes that AI can make mistakes and calls for close supervision and checks; undetected defects before production can cost millions. Hussain assures that Architect's AI system checks its own work to spot these errors rather than relying on humans.

The Startup Describes Redwood Designed and Verified in Two Weeks

Architect Labs claims to have designed and verified a chip in two weeks thanks to AI. The design, dubbed Redwood, was reportedly completed within this timeframe. Two human architects framed and refined the overall architecture while AI handled the detailed design. An AI model executed on Redwood then identified possible circuit improvements, which, according to Ebrahim Hussain, sketches a self-reinforcing cycle. He states that he wants to develop larger models to allow AI to improve recursively and produce increasingly powerful hardware.

FPGA Tests Suggest Superior Efficiency to Jetson Orin Nano, According to Architect

At this stage, the design has been evaluated on reprogrammable FPGA hardware. Architect Labs claims that these tests suggest the finished product could be more efficient than Nvidia's Jetson Orin Nano, which is used as an AI supercomputer in robotics. The company also states that Redwood can run models such as Meta's Llama and Alibaba's Qwen.

According to Architect, Designing a Chip Takes a Year or More and About 100 People

According to Architect Labs, designing a chip typically takes a year or more, costs hundreds of millions of dollars, and requires specialized teams of about 100 people. Ebrahim Hussain believes this time lag is problematic for AI, whose models evolve faster than chips. Hao Zheng finds the claimed timeline for Redwood impressive, while noting that several startups and established players like Synopsys are already using AI to accelerate design. The discipline remains one of the most demanding in the tech industry.

$24M Raised in June and Ambition Displayed Towards a "Designless" Era

Founded last year, Architect Labs emerged from the shadows in June with a $24 million funding round led by Kindred Ventures. Jeff Dean, former chief scientist at Google, along with leaders from OpenAI and Nvidia, are among the supporters. The startup claims to bet on artificial intelligence to shorten design timelines, including for companies seeking custom hardware. Ebrahim Hussain, 20, believes AI could usher in a new era: after the "fabless" age in which TSMC enabled players like Nvidia to develop chips without owning factories, he considers the next step will be "designless," allowing access to custom chips without the need for the large teams typically required.