The semiconductor industry closed 2023 on a high note as the U.S. CHIPS Act brought renewed attention to our importance and product introductions from startups and established players buoyed our confidence.
Technological innovations will continue into 2024 along the same trends line much as they did last year more or less. Despite a drop in overall semiconductor revenues of about 10%, top billing for 2023 tech innovation from semiconductor industry followers would have to be OpenAI, Generative AI and ChatGPT, as we witnessed enormous interest in GenAI once ChatGPT took center stage. Following closely behind would have to be the accelerated adoption of RISC-V and the embracing of chiplets for the promise of bending the Moore’s Law curve. The incoming year will also see renewed interest in autonomous driving (AD) at Level 4 where vehicles can operate in self-driving mode and in fully autonomous vehicles at Level 5 that no longer require human attention behind a steering wheel. In fact, steering wheels will not be a standard feature.
Beside a projected increase of global semiconductor sales in the ballpark of 17%, 2024 will usher in some challenges. Notwithstanding the drawbacks of massive computational power and outsize energy consumption, GenAI will continue to revolutionize all sectors driven by human behaviors in every industry from commerce, agriculture and finance to medicine, pharmacy and entertainment, and its adoption will expand exponentially.
Still, for all the progress in processing power that in 2023 broke the celling of multiple petaflops, the inherent potential of AI and GenAI is severely incumbered by the limited memory bandwidth of memories. Although GPUs are the top choice for AI learning in data centers, their efficiencies are purportedly in the range of 1% to 9%, which is abysmally low. Semiconductor companies have observed this limitation and some are ambitious enough to take up the challenge to accelerate adoption in edge applications, which includes the largest markets today.
What is needed are novel chip architectures and processing methods, including robust and tightly coupled software stacks. Breakthrough architectures must address memory and data transfer bottlenecks, and drastically enhance processing efficiencies while reducing latencies. Likewise, focus should be directed to limit power consumption, a roadblock to support inference of GenAI at the edge and on mobile devices.
To work around the need for massive compute power to execute large language models and avoid the one-size-fits-all approach, the industry will develop domain-specific AI/ML customizations. Smaller chip designs targeting only one application can be accommodated by existing semiconductor architectures with minor enhancements. Early signs of this new trend popped up in 2023. The trend will continue in 2024 with broad acceptance.
At the closing of 2023, the consulting company SHD Group delivered a RISC-V market analysis with a prediction of revenue growth from RISC-V based SoCs, including royalties, licensing, maintenance and support, from about $100 million in 2023 to about $250 million in 2024 and almost $100 billion in 2023.
In the course of 2023, chiplets attracted lots of attention. In 2024, new AI/ML large designs under development will make chiplets a viable alternative to build huge monolithic chips at the most advanced and expensive process technologies.
In 2024, the investment community will re-open its deep pockets and pour cash in pioneering semiconductor startups with daring technology inventions and aggressive business plans.
About Lauro Rizzatti
Lauro Rizzatti is a business advisor to VSORA, an innovative startup offering silicon IP solutions and silicon chips, and a noted verification consultant and industry expert on hardware emulation. Previously, he held positions in management, product marketing, technical marketing and engineering.