Can Computer Science Students Be Taught to Design Hardware?

The semiconductor industry is grappling with a significant talent shortage, prompting experts to explore whether computer science students can be trained to design hardware. This question has gained urgency as the demand for chip designers continues to outpace the supply of electrical engineering graduates.

The Talent Gap Challenge

The chip industry faces a stark imbalance in workforce demographics. According to Jason Cong, distinguished professor of computer science at UCLA, there are approximately 2 million software developers in the United States compared to fewer than 100,000 hardware designers. This 20:1 ratio highlights the potential opportunity in cross-training software engineers for hardware design roles.

AI as an Enabling Technology

Artificial intelligence is emerging as a powerful tool to bridge the skills gap. New AI-powered design tools can automate many low-level details of hardware development, from generating testbenches to optimizing layouts and suggesting design improvements. These tools make it easier for those without deep hardware expertise to contribute meaningfully to chip design.

Matthew Graham, senior group director at Cadence, explains that future chip developers will have different skills than today's engineers. "It will be closer to what a software engineer does," he said. "Will we exclusively take software engineers and not train them at all, and they'll be able to generate hardware? I doubt it. There's still some specific domain knowledge that's required."

Educational Innovation

Universities are experimenting with various approaches to address the talent shortage. Cong has been teaching an undergraduate course where students learn to design CNN accelerators using high-level synthesis in just one-and-a-half weeks. His goal is to make hardware design "just as easy as writing another PyTorch library."

The approach involves a multi-agent-based system that combines machine and human intelligence. This framework captures design hierarchy, enables program transformations, and facilitates task transfer between different hardware platforms.

The Role of Agentic AI

Agentic AI systems are proving particularly valuable in this transition. These systems can provide conversational interfaces that guide users through complex design tasks without requiring deep expertise in hardware description languages like SystemVerilog or VHDL. Sathishkumar Balasubramanian, head of products at Siemens EDA, notes that these tools become more effective as they're fed more data about previous design work and projects.

Industry Perspectives

Companies are taking different approaches to hiring and training. Alexander Petr, senior director at Keysight EDA, observes that the industry is following two trends: hiring computer science graduates with basic electrical engineering understanding and training them, or upskilling electrical engineers on the computer science side.

Startups formed from university research have an advantage in accessing talent. ChipAgents, founded by AI professor William Wang at UC Santa Barbara, was able to hire 80-90% of its team from the university. This model has precedent, with companies like Broadcom, Qualcomm, and Cadence originating from academic research.

The Future of Engineering Education

While some organizations are exploring shortened degree programs, experts suggest that reducing study time may not be as beneficial as doing more within the existing timeframe. Andrew Johnson from Imagination Technologies notes that the drive to reduce class time from three years to two years is interesting, but questions whether it's better to stick with three years and do more in that time.

Practical Considerations

Despite advances in AI tools, human expertise remains essential. Andy Nightingale, vice president at Arteris, emphasizes that "somebody needs to be there as the quality check, the quality gate, before things get committed." The fear shouldn't be AI replacing roles, but rather people who know how to effectively use AI tools replacing those who don't.

Conclusion

The evidence suggests that computer science students can indeed be taught to design hardware, but it requires a combination of AI tools, educational innovation, and recognition that some domain-specific knowledge remains essential. The key is not to replace electrical engineers but to expand the pool of people who can contribute to hardware design through new tools and approaches.

The semiconductor industry's future may well depend on successfully bridging the gap between software and hardware expertise, creating a new generation of engineers who can work effectively across both domains.