- Intel and Altera are building modular reference designs for physical AI that combine compute, software and chips
- FPGAs let developers support many sensor types by reprogramming the chip instead of changing hardware
- The designs target healthcare, aerospace and defense, and robotics, in a physical AI market analyst Jack Gold estimates at upwards of $50 billion in the coming years
When robots roam the world, they will rely on sensors and compute power. But it turns out passing the data from one to the other is easier said than done. And that’s exactly the problem Intel is trying to solve with some help from Altera.
The pair announced they’re working on a series of modular reference designs that will combine Intel’s compute, connectivity and software with Altera’s Field-Programmable Gate Arrays (FPGAs), a type of chip that can notably be re-programmed after it is manufactured. Here’s why that matters.
Jack Gold, founder and principal at J. Gold Associates, explained in a note to investors that sensors – like the ones that physical AI will rely on – are “highly varied in protocol and data formats.” That can make things tricky when different sensors need to interact with the same base system.
“Having an FPGA that can be programmed based on the type of sensor and the format of data it produces is a critical capability in designing Physical AI systems. And it further allows a wide range of sensors to be used by simply reprogramming the FPGA without a need to change the physical hardware,” he wrote. “At this time, only AMD has similar in-house FPGA capabilities to the Intel Altera collaboration.”
What Intel and Altera are building
Gold noted that Intel here is extending the same reference design strategy is used for PCs in the physical AI realm to help speed development.
Specifically, Intel and Altera are building reference designs targeting healthcare (with an emphasis on low-latency), aerospace and defense (with a focus on size, weight and power), and robotics (zeroing in on real-time processing requirements).
“As sensors multiply at the edge and through Physical AI systems, our customers need a simpler way to turn real-world data into timely action,” Dan Rodriguez, corporate VP and GM for Intel’s Edge Computing Group, said in a statement. “Together with Altera, we aim to give OEMs a flexible foundation that simplifies integration and helps them move faster from prototype to market deployment while meeting their needs across performance, power and cost.”
Why it matters
Gold estimated that the market for physical AI will grow to between $50 billion and $65 billion in the next few years, with components like processors, sensors and software making up as much as 50% of that total. He added Intel’s partnership with Altera will likely help its competitive positioning as it squares off with the likes of Nivida, Qualcomm, Mediatek, Broadcom and others in this market.
Of course, sensors and compute aren’t all that will matter in a physical AI world. Ericsson, Nokia, and other telecom vendors are positioning the network as a critical foundational layer for physical AI.
“Any system that moves – so drones, vehicles, humanoids, robots talking to other robots…all these will become an autonomous system that needs a network to think with it,” Ericsson Americas Chief Strategy and Technology Officer Joe Constantine told Fierce in August.
Read more about physical AI on Fierce Network:
Schneider teams with HPE and Nvidia on industrial AI agents at the edge
Opinion: Physical AI doesn’t need low latency
Robotics could give telcos their best shot at winning in AI
Ericsson: Telco edge was early, not wrong — AI may finally make it real