Renesas Pioneers Real‑Time Continuous AI for Factory Automation

TOKYO — In the age of smart manufacturing, artificial intelligence is no longer optional; it’s a cornerstone. Yet many firms struggle to translate AI theory into operational reality, citing a lack of in‑house expertise and uncertain ROI. Renesas Electronics is redefining the path forward with its “real‑time continuous AI” vision for operational technology (OT).

Unlike the data‑driven, cloud‑centric “statistical AI” popular in IT, Renesas’s approach embeds AI directly at factory endpoints. This allows machines to detect anomalies and make decisions instantly, without round‑trip latency to the cloud.

click for larger image

Statistical AI for IT vs. continuous AI for OT (Source: Renesas)

Yoshikazu Yokota, executive vice president and general manager of Renesas’s industrial solution unit, explained to EE Times that embedded AI is critical for fault detection and predictive maintenance. “When an anomaly arises, embedded AI can respond locally and in real time,” he said. Three years ago, Renesas first proposed “AI at endpoints” and tested the concept in its own Naka semiconductor fab.

“Our goal is to enable real‑time inference in OT while incrementally expanding AI capabilities at the edge,” Yokota added.

Yoshikazu Yokota, executive vice president at Renesas, plans to focus on offering real‑time inference in OT. (Photo: EE Times)

By introducing AI in incremental steps, Renesas aims to help manufacturers complete proof‑of‑concept projects more quickly and understand the true ROI of their AI investments.

When to Apply AI to OT

Mitsuo Baba, senior director of strategy and planning for Renesas’s Industrial Solution unit, emphasized that AI delivers maximum value when it targets well‑defined problems—such as specific production line anomalies. He illustrated the concept with a scenario: a seasoned operations manager can use AI to flag the exact moment a defect begins, freeing the manager to focus on critical interventions rather than routine inspections.

In this model, AI is trained once on identified issues and then performs inference on endpoint devices in real time, eliminating the need to send data to the cloud. Baba highlighted that only 30 KB of data is sufficient for edge inference, compared to the 300 MB typical of cloud‑based statistical AI.

Renesas is therefore championing an inference‑only approach that can run on microcontroller units (MCUs), allowing existing production equipment to be retrofitted with an “AI Unit Solution” kit rather than overhauling entire lines.

“We’re not competing with AI chip leaders like Nvidia,” Baba clarified. “Our mission is to pioneer a new segment of embedded AI where the data footprint is small enough to run on conventional MCU/MPU hardware.”