Sailesh Chittipeddi | Government Vice President Operations | Renesas
Digital transformation has developed shortly from a conceptual section to a semiconductor trade change agent. The fast take up of AI-enhanced product improvement is barely accelerating this transformation and is additional influenced by two related developments: The motion of Moore’s Legislation from transistor scaling to system-level scaling, and the comparatively current redistribution of the worldwide electronics provide chain as a result of COVID-19 pandemic.
I spoke on this topic earlier this month on the Trade Technique Symposium 2024 in Half Moon Bay, California, the place leaders from throughout the chip trade collect yearly to share their insights on expertise and development drivers and what they might imply for our respective companies.
Between the early Nineteen Seventies and round 2005, elevated chip efficiency was largely a perform of clock frequency enhancements pushed by advances in lithography, transistor density, and power effectivity. With growing transistor counts (and die dimension), clock frequencies are restricted by interconnect delays and never by transistor efficiency. To beat this problem, designers moved to multi-core designs to extend system efficiency with out blowing up power. Novel packaging strategies akin to chiplets and multi-chip modules are serving to additional enhance system efficiency, significantly in AI chips.
A single chip package deal could also be comprised of a number of chiplets every housing particular features akin to high-performance logic components, AI accelerators, high-bandwidth DDR reminiscence, and high-speed peripherals. Fairly often, every of those elements is sourced from a distinct fab, a development that has resulted in a fragmented world provide chain. This creates its personal set of challenges as die from a number of fabs should be built-in right into a package deal or system that should then be completely examined. Check failures at this stage have huge monetary penalties. These challenges, require a “shift left” mindset in product improvement. The shift left mentality has main ramifications for a way we, as an trade, must be managing our provide chains by transferring the heavy emphasis from structure/design to ultimate system testing and high quality.
Provide chain challenges throughout the COVID pandemic have resulted in additional decentralization of the provision chain elements. As an instance the enormity of the change underway, contemplate that between 2022 and December 2024 building started on 93 wafer fabs all over the world. Evaluate that to the worldwide building of automated take a look at amenities. In 2021 alone, the trade broke floor on 484 back-end take a look at websites, which gives a measure of how dedicated the chip sector is to driving resiliency throughout the manufacturing panorama.
The Function of AI in Semiconductor Design and Manufacture
So, the place does AI come into the image?
A key space by which AI will exert its affect is the shift from an analytic to a predictive mannequin. Right this moment, we wait to detect an issue after which have a look at previous information to determine the foundation explanation for the issue and stop it from reoccurring. This inefficient method provides time, price, unpredictability, and waste to the provision chain. AI, however, permits us to look at present information to foretell future outcomes.
As a substitute of utilizing spreadsheets to investigate previous information, we construct AI fashions that manufacturing engineers constantly prepare with new information. This “new” information is not merely a set of numbers or measurements however consists of unstructured information akin to die pictures, gear noise, time sequence sensor information, and movies to make higher predictions.
In the long run, it’s about pulling actionable data from a sea of knowledge factors. In different phrases, information with out motion is generally ineffective. Why am I driving this level dwelling? As a result of at this time, 90 % of knowledge created by enterprises isn’t used. It’s darkish information. And when you concentrate on AI implementation, 46 % of them by no means make it from pilot to manufacturing as a result of the complexity of the packages isn’t scoped appropriately.
Regardless of these challenges, gear makers are already beginning to implement digital transformation strategies into their product improvement processes. The advantages are palpable. Analysis from Boston Consulting Group discovered that firms which have constructed resiliency into their provide and design chains recovered from COVID-related downturns twice as quick as firms which have but to embrace digital transformation.
At Renesas, we acquired an organization known as Actuality AI that generates a compact machine studying mannequin that runs on a microcontroller or microprocessor. This gives the distinctive capability to shortly detect deviations from regular patterns that will trigger gear issues. It permits manufacturing amenities to schedule preventive upkeep or reduce downtime related to sudden gear failure.
Digital Transformation Is Future-Proofing Our Trade
Digital transformation with AI is vital to enterprise success at this time. Because the semiconductor trade undergoes a significant evolution – embracing system-level design and adapting to a altering world provide chain – digital transformation and the shift left method are highly effective instruments that ship on two fronts.
The primary is a productiveness improve that comes from optimized instruments and design processes. The nearer you’re to the place the failure is prone to happen, the extra shortly you be taught and the extra shortly you possibly can make things better.
Second, and maybe most significantly, digital transformation solves one of many largest issues the trade has with chip design – the supply of expertise. After we scale back the time taken to design a chip, we’re making our engineers much more environment friendly than they’d be in any other case, which is more and more necessary because the semiconductor trade demographic skews older.