The rapid evolution of silicon chips has significantly propelled the advancement of artificial intelligence (AI). In a promising twist, Cognichip is leveraging AI to innovate the chip design process, which has long been plagued by complexity, high costs, and lengthy timelines. Traditionally, developing advanced chips can take anywhere from three to five years, with the design phase alone sometimes extending to two years, as exemplified by Nvidia’s latest GPUs, which feature a staggering 104 billion transistors.
Cognichip’s CEO, Faraj Aalaei, highlights that the lengthy timeframe to develop a chip can lead to market shifts, potentially rendering significant investments obsolete. The company’s mission is to integrate AI tools commonly used by software engineers into semiconductor design to expedite the process. Aalaei asserts that these advanced systems can produce efficient, high-quality code when properly guided, potentially reducing chip development costs by over 75% and halving the time required.
Having emerged from stealth mode last year, Cognichip announced it has secured $60 million in funding from Seligman Ventures, with additional investments from notable figures in the industry, including Intel CEO Lip-Bu Tan. The total funding since the company was founded in 2024 has reached $93 million. However, it has yet to reveal any chips developed using its system or disclose its collaborative customers since September.
Cognichip distinguishes itself by utilising a proprietary model trained specifically on chip design data rather than general-purpose AI models. Acquiring domain-specific training data has proven challenging, as chip designers typically keep their intellectual property closely guarded, in stark contrast to the open-source environment favoured by software developers. To overcome this hurdle, Cognichip has created synthetic datasets and licensed data from partners, enabling secure training on proprietary data without risk of exposure.
Where proprietary datasets are lacking, Cognichip has called upon open-source alternatives. During a hackathon last year, it demonstrated its model by allowing electrical engineering students at San Jose State University to design CPUs based on the RISC-V open-source architecture.
In a competitive landscape, Cognichip faces established firms such as Synopsys and Cadence Design Systems, as well as several well-capitalised startups, including Alpha Design AI and ChipAgentsAI. Umesh Padval of Seligman Ventures remarked that the current influx of investments into AI infrastructure is unprecedented in his 40 years of experience, suggesting that we may be entering a transformative period not only for the semiconductor sector but also for innovative companies like Cognichip.
Fanpage:Â TechArena.au
Watch more about AI – Artificial Intelligence
