Can There Be Any Doubt? In Our Industry, 2023 Was ‘The Year of AI’

Usually when we do these end–of-year top trends articles, we examine four or five significant trends in the electronics and semiconductor industries. There’s usually more than one top trend to discuss. But 2023 has been different. There’s nothing that has so infused every corner of electronic design this year more than artificial intelligence (AI).

In the culture in general generative AI, including ChatGPT, very much dominated this year’s buzz. But, in our industry, AI has loomed large as well.
 

At CES 2023, AMD Chair and CEO Dr. Lisa Su declared, “AI is the defining megatrend in technology.”

 

It’s true that AI was a huge part of this year’s mindshare, and you saw it in a lot of our headlines on our site. But it’s interesting to note that AI was actually not a significant part of our highest page view articles. Look for an article rounding up our top 2023 All About Circuits news stories on New Year’s Eve. Perhaps that shows that engineers like you have information preferences that stretch beyond that of AI. AI is real, important, and here to stay, but it's just one tool in an engineer's toolbox.

While AI was the dominant technology trend this year, there were definitely subcategories in AI technology advances relevant to engineering. In this article, we break these down and examine each one.

• AI in processing and accelerators
• AI processing aimed the the edge and application-specific task
• AI in PCB design tools, IC design tools, and embedded development tools

 

AI in Desktop Processing, Accelerator Subsystems, and More

For its part, AMD kicked off 2023 with bold moves in AI by announcing at CES 2023 the industry’s first x86 processor with dedicated AI hardware. This took the form of what AMD calls its XDNA adaptive AI architecture.

The scheme enables an ability to pass large arrays of AI data from layer to layer. This approach is said to be customizable to whatever AI workload you need to process. AMD’s 2023 announcements included a desktop x86 processor that embeds an AI hardware engine that uses XDNA technology.

 

Intel 5th Gen Xeon processors with AI acceleration built in. Image used courtesy of Intel

 

Meanwhile, Intel saved its AI x86 rollouts for later in the year. At Hot Chips 2023 in August, the company rolled out two new Intel Xeon processors. Codenamed Sierra Forest and Granite Rapids, the devices use new architectures to access higher performance and flexibility. For instance, the Granite Rapids chip claims to offer a 2–3x performance boost for AI workloads.

Intel's most significant AI move came only a couple of weeks ago, with Intel announcing its 5th Gen Xeon CPUs and Core Ultra mobile processors, both with built-in AI acceleration. The company called these efforts as part of a plan to enable AI processing both in the cloud and at the edge.

 

Several Companies Up Their AI Game

Intel and AMD were far from alone amongst big companies weighing in with AI processing solutions. Both rolled out in November, examples include Nvidia’s H200 Tensor Core GPU and Microsoft’s Maia AI Accelerator and Azure Cobalt CPU. For its part, Qualcomm targeted AI for IoT with a family of processors it announced back in April.

 

Nvidia H200 Tensor Core GPU. Image used courtesy of Nvidia

 

Other interesting AI processing announcements included SiFive’s October release of RISC-V cores aimed at generative AI and machine learning (ML). Start-up Tenstorrent emerged on the scene likewise using RISC-V as the basis for AI processing. Meanwhile, IBM this summer weighed in with an analog AI chip that marries analog in-memory computing with scalable digital interfaces to realize AI building blocks.

Eyeing AI workloads in the data center is Korean startup Sapeon, who this fall rolled out its X330 AI processor. Ambarella focused on security cameras with its CV72S chip, claiming it as the security industry’s highest AI performance-per-Watt SoC.

 

MCUs Vendors and Startups Eye AI at the Edge

When the internet-of-things (IoT) came into existence as a trend over a decade ago, the general consensus was that AI computing would have to be done in the cloud. Fast forward to today, and clearly there are a myriad of solutions available to do AI processing at the edge. These have emerged from both traditional microcontroller (MCU) vendors and also from several AI chip startups, each offering unique solutions.

Arm cores have become dominant among MCU cores these days. The company released its Cortex-M52 core back in April, positioning it as an enabler for adding AI/ML into mainstream embedded devices. Just this month, Renesas unveiled an Arm Cortex-M85 based MCU design for graphics and multimodal AI.

 

The PSoC Edge family of MCUs offers built-in ML support, Autonomous Analog, and many more features that make it a candidate for next-generation smart edge devices. Image used courtesy of Infineon

 

Also, from the MCU space this fall, Infineon launched its PSoC Edge family of MCUs. Aside from Cortex-M55-enabled DSP support and an Ethos-U55 NPU, the PSoC Edge devices sport Infineon’s proprietary neural network hardware accelerator NNLite.

Startups in the MCU space likewise had AI in their sights. Case in point, Alif Semiconductor launched its Ensemble line of MCUs that combine a computing core with ML acceleration to enable heavy ML workloads in battery-operated devices.

 

AI Startup Success

2023 was definitely a big year for startups in the AI chip space. Hailo expanded its Hailo-8 family, adding its Hailo-8 Century that brings PCIe support to the Hailo-8 line, and Hailo-8L, an entry-level solution for AI acceleration. Kinara rolled out its tiny 17 mm x 17 mm Ara-2 processor designed for large generative AI workloads.
 

In a tiny 17 mm × 17 mm EHS-FCBGA package, the Ara-2 chip is designed around 8 Gen-2 neural cores. image used courtesy of Kinara

 

Backed by Nvidia, AI-networking startup Enfabrica this year opened pre-orders for its 8 Terabit/s ACF-S switch system based on its technology. Finally, Axelera had a big funding year while pushing the in-memory computing approach of its Metis processor.

 

AI Infiltrates Tools for PCB Design, IC Design, and Embedded Development

There’s no doubt that 2023 was a huge year for AI—including generative AI—in a variety of hardware and software design tools. Siemens was first up back in February with its Questa Verification IQ, a software platform designed to use AI for data-driven IC verification.

The Spring of this year saw AI-based releases from both Synopsys and Cadence. Synopsys rolled out Synopsys.ai, a suite of AI-driven tools for the design, verification, testing, and manufacturing of advanced ICs. Meanwhile, Cadence launched Allegro X AI technology aimed at automating placement and routing using AI.
 

Cadence’s Allegro X AI inputs PCB design information such as netlist and physical constraints, and outputs board placement, critical routing, and more. Image used courtesy of Cadence

 

The Spring also saw the launch of Flux's Copilot, which is essentially an AI-based design assistant integrated into a PCB design tool. More recently, the company this month unveiled a new version of Copilot that provides the capability to recognize and work with images like block diagrams.

On the embedded software development side, STMicroelectronics (ST) brought its free STM32 MCU Edge AI toolset to the cloud back in January. And just this month, the company announced (also free) new AI software libraries for unlimited deployment on any STM32 MCU. In a hardware/software combo announcement, Ambarella co-developed a neural network processing stack and chip family for AI computing in autonomous vehicles.

 

More AI Everywhere in 2024

I have to be honest. Despite all the epic AI news stories All About Circuits covered this year, this one is probably my favorite. Our own Jake Hertz got this exclusive interview with researchers from NYU. They successfully used ChatGPT to design and build a microprocessor. Jake had the chance to meet NYU professor Dr. Hammond Pearce to learn more about this research face to face.

While AI was certainly big in our industry in 2023, there is little doubt that 2024 will see more of the same. The ability for AI and ML to speed up all manner of engineering and design tasks makes them irresistible technologies. Chip and tool vendors—both big and small, both old and new—will find new ways to use AI and to improve on how they’re already using it.