Aizip and Renesas have demonstrated ultra-efficient small language models and compact AI agents on Arm-based micro-processor units (MPUs) for applications on edge systems. This could pave the way for human-AI interactions in home appliances, enterprise kiosks, retail and industrial appliances and other edge devices.

Small language models (SLM) enable AI agents on device edge applications providing the same features of large language models (LLMs), but on edge footprints. On-device models offer advantages that include enhanced privacy protection, resilient operation, and cost savings. While several companies have successfully reduced the size of language models for mobile phones, ensuring accurate tool calling for automation applications on low-cost edge devices remains a significant challenge for these SLMs.

Aizip aims to enable pervasive intelligence by building ultra-efficient, robust, and scalable AI models that can be deployed anywhere. The company has created a series of ultra-efficient small language models (SLMs) and AI agents, named Gizmo, ranging in size from 300 million to 2 billion parameters. These models support diverse platforms, including MPUs and application processors for a broad range of applications.

Aizip has teamed up with Renesas and Arm to demonstrate compact and efficient AI agents on MPUs. These models have been deployed on Renesas RZ/G2L and RZ/G3S boards. Built on Arm Cortex-A55, the Renesas MPUs are now available for commercial use.

Despite their compactness, these SLMs and SLM-based AI agents achieve robustness and accuracy comparable with cloud-based large models for domain-specific applications. Employing an AI design automation pipeline, these SLMs and SLM-based AI agents can be rapidly generated for new applications.

“We’re witnessing a pivotal moment in efficient AI where on-device AI agents can truly make everything think,” described Weier Wan, a founding member of Aizip and head of Aizip’s SLM and AI-agent development. According to Weier, on the RZ/G2L with a single A55 core running at 1.2 GHz, they can achieve a response time of less than 3 seconds. “With Renesas’s readily available IC, we can offer the AI SLM and agents right now. Renesas devices are known for their high quality and mature development environment, and we’re excited to team up with them in offering this breakthrough solution.”

Renesas, a provider of MPUs and MCUs, has been developing technologies for edge and endpoint markets. “We see AI moving rapidly in our direction, and we are ready for it,” said Kaushal Vora, Sr. Director of Business Acceleration and Ecosystem of Renesas. “We are particularly excited about Aizip implementing an on-device agent on our MPU. We can envision, for example, RealityCheck HVAC could predict a potential issue with a component in a heat pump; the AI agent acts on this real-time operational information, looking up the appropriate OEM part and scheduling a service call with the technician for replacement. Aizip has been an excellent strategic partner, and we have achieved multiple milestones together,” Vora further commented.

“To address the rapid growth of AIoT applications, devices require higher performance and intelligence at the edge,” said Lionel Belnet, Senior Director HW Product Management, IoT Line of Business at Arm. “Arm is the platform of choice for AI solutions from cloud to edge, and running Aizip’s new SLMs and AI Agents on Arm-based technologies will provide the performance and efficiency needed to scale the edge AI opportunity even further.”

“Efficiency is critical for AI systems,” said Yubei Chen, co-founder of Aizip and assistant professor at UC Davis. “Natural intelligence demonstrates remarkable efficiency. From a scientific perspective, there’s immense potential for AI to improve, and efficiency is the key to achieving human-level intelligence.” The Aizip Intelligent Language (AIL) model series, including AI agents, is an extension of Aizip’s efforts to bring efficiency and sustainability to real-world applications. “Aizip is excited to be a part of the Arm ecosystem, utilising industry-leading design solutions to deliver new AI use cases worldwide.”

The partnership between Aizip and Renesas has led to tinyML models in audio, vision, and time-series. They are now collaborating in the market of SLMs and agents for on-device applications.