Edge AI will shape all aspects of MediaTek business

Misha Lu, DIGITIMES Asia

TAIPEI, June 27, 2023 /PRNewswire/ — The global AI market is expanding at a rapid and unprecedented pace, transforming the semiconductor industry at the same time. As the demand for high-performance chips increases with the widening scope of AI applications, DIGITIMES Asia, in cooperation with Cadence and MediaTek, is set to host the 2023 Strategy+ Symposium @ Bay Area on July 19th, 2023, aptly titled ‘AI Ushers in the Golden age of Semiconductor industry’, to explore how the latest AI revolution changes the IC design sector in the context of generative AI, edge computing and Asia’s supply chain.

Dr. Finbarr Moynihan, head of Corporate Marketing at MediaTek, will speak at the event to explore AI adoption for smart edge devices. In a preliminary interview with DIGITIMES Asia, the MediaTek VP shares his insights on where the leading mobile chipmaker stands when it comes to AI deployment, and what AI means for the IC design sector.

For many years, according to Moynihan, MediaTek has been integrating AI capabilities into several chipsets for different industry segments. However, early deployments a few years ago were mostly software-based, leveraging existing CPU and GPU architectures on-device. Early focuses were mostly on applications like voice assistants and smart speakers.

Over time, MediaTek developed custom AI Processing Units (APU) to offer more power efficient hardware solutions that accelerate more common AI capabilities on devices, including the support for image, video and gaming applications. Other applications include digital TV solutions for AI picture quality and AI Super Resolution. “As the AI applications continue to multiply, we are adding more and more APU capabilities to most of our SoCs,” indicated the MediaTek VP.

Edge AI applications, according to Moynihan, will likely shape all aspects of MediaTek business. “For sure most, if not all, of our SoC solutions for smartphones, digital TV, automobile, industrial and IoT, smart speakers & displays, Chromebooks, tablets and smart home will continue to integrate more and more capabilities for AI applications at the edge,” he indicated. Due to cost, power, data bandwidth and privacy concerns, Moynihan believes that AI will require strong on-device capabilities since it will not be possible to execute all of this in the cloud. Especially, with the rise of generative AI, the demands on edge computing will dramatically increase. “We are just on the cusp of this becoming a reality, but more work will be needed,” noted Moynihan.

The MediaTek VP sees generative AI enabling the deployment of much more capable and personalized use cases on device, but pointed out the key prerequisites for edge AI to be of practical use, such as having significant on-chip and low-power acceleration capability. In addition, these techniques will also push the need for more and more memory on devices.

Meanwhile, the AI revolution underway is also changing the IC design and manufacturing process itself. In addition to applying AI techniques to help optimize the task associated with complex SoC system design and layout, AI can also help in yield enhancement and manufacturing improvements for production at high volumes, according to Moynihan, who finds these AI techniques especially useful for rapid deployment of advanced SoC for fast moving consumer markets.

AI deployment in IC design is not without challenges though. For the MediaTek’s VP, a major obstacle is the fact that the development of AI applications often moves at a pace faster than the development time of advanced SoC solutions based on leading edge process technology. Such pace makes it harder to predict the appropriate balance between hardware and power-optimized circuits (e.g. APU) versus more generic and software flexible solutions (such as CPU and GPU) to include in a SoC design. In addition, Moynihan pointed out that there are a lot of fragmentations and architectures on the software side of AI development – namely the platforms used by developers – that need to be managed to deliver a scalable, flexible and usable AI platform.

Register and join us at the symposium on July 19th, 2023 to be part of this pioneering discussion.

For media enquires, please contact: joyce.siow@digitimes.com