How many transistors can you fit on a tiny piece of silicon, an unassuming platform that holds a vast number of integrated circuits (ICs) that power technology all over the world? In 1971, the first microprocessor was built with 2,300 transistors – but today’s silicon can boast more than 100 billion. Before the decline of Moore’s Law, the number of transistors that could be packed onto an IC was roughly doubling every two years, and the Holy Grail of chip design was to get as small as possible — but we hit a wall with the laws of physics. Today’s smallest practical transistors measure just two nanometers (nm) across, a little larger than your typical atom.
As we try to scale that wall and continue to innovate in chip design, I see several trends emerging that we should watch in the semiconductor industry. Some, like cloud, have been developing for years and are becoming as normalized as traditional data center computing, and others, like artificial intelligence (AI) and machine learning, are still in their infancy but penetrating nearly every area of human technology development.
Semiconductor Design Trend #1: Reaching for the Sky
These days, instead of making transistors smaller and packing them more densely, IC designers are building up. They’re getting architectural with multi-layer structures and stacking the newest generation of semiconductors. Like city planners, when IC designers run out of horizontal space to build, the natural solution is to expand upward.
Vertical stacking can give 3D ICs additional functionality, reduced form factor, and improved interconnect density — but stacking can also introduce challenges, including thermal management. Stacking-induced thermal stress — excessive heat — can cause performance problems and mechanical failures, ultimately compromising an organization’s productivity, product quality, and time to market. CPU cooling is an issue in any system and is especially pronounced when transistors are stacked with less surface area exposed. Because Altair understands the challenges faced by today’s semiconductor industry, our solutions can help designers optimize thermal management strategies and facilitate the implementation of 3D IC technologies in advanced electronic systems.
Semiconductor Design Trend #2: Bringing Design In-House
Beginning with the COVID-19 lockdowns of 2020 and demand for work-from-home technology, a worldwide microchip shortage has compelled companies — including major manufacturers in industries like automotive and aerospace — to turn in-house and design their own ICs. While most have traditionally outsourced the processors that power their technology, a growing number of companies have now found they need take chip design into their own hands. While the shortage is easing, it has already changed critical processes in many organizations. They’re using electronic design automation (EDA) tools more than ever before, and because those tools can be very expensive, efficient resource orchestration is critical to ensure licenses are optimized, workloads run smoothly, and users can stay productive without waiting in long queues or struggling with IT headaches. In addition to EDA tools, some manufacturers are also using hardware emulation along with Altair solutions for hardware emulation environments. We deliver a complete suite of solutions for semiconductor design and EDA.
As the semiconductor design industry gets back to business as usual, it remains to be seen whether designers and manufacturers will continue to benefit from in-house IC design or will find it more beneficial to return to outsourcing. Since they’ve already invested a significant amount of time and money in their own processes and equipment, many may prefer to retain the benefits of control over quality and production time.
Semiconductor Design Trend #3: Designing in the Cloud
Cloud is a long-term trend that’s been steadily building as the technology to support efficient cloud computing gets faster, cheaper, and more advanced, removing barriers to entry like high cost and steep user learning curves. Today’s semiconductor designers aren’t just building up their silicon chips and bringing IC design in-house, they’re also elevating their computing into the cloud. With more EDA users than ever before using cloud and hybrid cloud for its virtually unlimited capacity and low cost of entry, effectively managing computing resources is crucial and ensures productivity stays high and costs stay low.
À la carte cloud computing can be an attractive alternative to making a large upfront investment in data center hardware and the expertise to run it, as has traditionally been the case with high-performance computing (HPC) and powerful on-premises supercomputers, often staffed by IT experts around the clock. Cloud compute cycles can add up quickly, though, and become expensive. According to the Semiconductor Industry Association, semiconductor companies in the U.S. invest around one-fifth of their revenue in research and development (R&D), which amounts to tens of billions of dollars each year. Altair’s HPC and cloud solutions enable companies to optimize resources, including expensive EDA tool licenses, and give users easy access and remote visualization on a unified, collaborative platform.
Semiconductor Design Trend #4: Artificial Intelligence and Machine Learning
Perhaps most critically, the pervasive global trend of AI and machine learning, which are working their way into every facet of life and changing the world as we know it in healthcare diagnostics, natural language generation, smart home control, and more — is impacting IC design and development. While AI is most visible in applications like ChatGPT that are generating a lot of buzz, the technology is also contributing in important ways to products and processes we already take for granted, like the silicon chips inside our computers.
Machine learning starts with training, in which large volumes of data are input and processed for later use in jobs like image identification and pattern recognition. Machine learning is already helping us reduce errors and increase precision in data analysis and decision-making, and while humans are still better at cognitive tasks like abstract analysis and subjective judgement, machine learning benefits from lack of bias and access to massive sets of hard data — the more the better.
EDA produces large volumes of data that are well-suited to machine learning, and designers are using machine learning to enhance EDA performance. With increasingly complex EDA tasks and the need for custom CPUs and GPUs with hundreds of cores and massive memory requirements, AI and machine learning are beating out traditional methods with less time and resource consumption, and companies are learning that using machine learning algorithms to improve IC design tools means they can work with higher levels of abstraction and deliver better results, faster.
Semiconductor Design Trend #5: Technology Convergence
What may be the most exciting trend of all is that we’ve been seeing convergence in nearly every area of enterprise computing, including the solutions we rely on for efficient semiconductor design. With the practical lines between HPC and high-throughput computing blurring, so much of it moving to the cloud, and decisions becoming increasingly data-driven, HPC and EDA are facilitating AI and machine learning; at the same time, AI and machine learning are boosting the power of HPC and EDA. Will all these branches of tech eventually converge until they’re indistinguishable, or will combined technologies take on lives of their own and branch further? Regardless, one thing is clear — AI and machine learning in EDA is the next big thing, and it's already here.
Convergence is leading to faster technology acceleration than we’ve ever seen and enabling innovations in semiconductor design, which has progressed from its humble roots five decades ago to performing on today’s leading edge. Despite challenges like growing complexity, chip shortages, global geopolitics, and whatever comes next, the integrated circuit continues to evolve and shape our technological future.