MIPS has today introduced its MIPS Atlas portfolio, a product suite comprising compute subsystems and software platform to develop autonomous edge solutions addressing the market opportunity for physical AI: the ability to sense, think and act in real time to deliver on the promise of precision intelligent industrial robotics and autonomous mobility platforms.
This development is significant both for MIPS in that it is a big move up the value chain, utilizing its experience from the automotive and advanced driver assistance systems (ADAS) data processing world and translating it into the bigger market opportunity from physical AI; more broadly, in terms of how silicon IP vendors obtain more value from the rapid growth in semiconductors where the fabless model is often described as not fit-for-purpose anymore.
In a briefing with EE Times, MIPS CEO Sameer Wasson highlighted how he was taking MIPS out of the IP licensing realm, moving the company from simply licensing processor cores to delivering much higher value sub-systems and silicon, with reference silicon available in 1H2026, and early pilot runs in 2027. “We are targeting the robotics and autonomous mobility market, building sub-systems and thinking beyond just semiconductor market,” he said.
In a blog released with the announcement, he explained, “The need for efficient autonomous platforms to advance next-generation driverless vehicles, factory automation, and many other applications is directly aligned with the MIPS Atlas portfolio. Our core competencies of safety, efficient data processing and experience in autonomy have enabled us to expand our portfolio with real-time intelligence that is the essential tech stack for physical AI platforms. MIPS customers can take our compute subsystems with software stacks as a turnkey solution to build physical AI platforms.”
He added that robots implementing physical AI will include agricultural tractors, self-driving warehouse robots and industrial manufacturing automation, healthcare, mobility assistance and self-driving vehicles. “Physical AI will enable greater flexibility, safety, precision, and reliability in the critical systems we rely on every day.”
Computing sub-systems
The MIPS Atlas portfolio subsystems comprise an application processor, compute cluster and microcontroller together with a software development platform. These aim to deliver application-specific turnkey enablers for the three categories of computing that make up physical AI: sense, think and act.
The premise of physical AI is to interpret surroundings from a diverse array of sensors that generate data, which in turn needs to be seamlessly transferred, integrated and processed in real time. The embedded AI engine of the Atlas portfolio processes this data to enable fast, private decision-making for safe and precise actions.
Wasson emphasized that none of the microcontrollers available today meet the need for smooth robot movements. Hence the M8500’s real-time compute platform capable of low-latency, control-loop processing in robotic movements that addresses this “action” part of the system, enabling dexterous control of motors and actuators.
Digital twin for hardware/software co-development
MIPS said its Atlas Explorer software platform enables its customers to analyze real-world workloads with deep insights into the micro-architectural function and efficiency, using high-fidelity processor modelling technology. The platform puts the workload first for hardware/software co-development, enabling a “shift-left” approach to system-level optimization. This means it helps enable digital twin conceptualization of production platforms for MIPS compute platforms, supporting long-term product life cycles.
The software development and optimization platform are said to combine the strengths of virtual platforms with micro-architectural performance modeling, enabling software teams to optimize workloads while hardware is still being designed. This approach provides insights allowing hardware teams to understand how real workloads perform, with direct visibility into workload execution rather than theoretical estimations. In addition, the Atlas Explorer software platform enables customers to employ “shift-left” methodology to develop and deploy their applications with MIPS Atlas subsystems in a digital twin universe.
MIPS said that its Atlas Explorer is delivered as a Visual Studio Code Extension, allowing developers to visualize performance data in a way that highlights critical optimization areas. Custom reports filters zoom into problem spots, making it easier to diagnose and resolve inefficiencies.
The company added that because performance data generation is decoupled from software development, even those outside the core team can contribute to optimization efforts without rebuilding software from scratch or migrating to different platforms.
Analyst firm HyperFRAME Research’s CEO, Steven Dickens, said the MIPS Atlas portfolio represented a “significant leap forward in enabling physical AI at the edge,” delivering the real-time compute power needed for autonomous platforms to thrive in industrial and automotive applications. “By integrating safety, efficiency and cutting-edge intelligence, MIPS is well-positioned to accelerate innovation across the rapidly expanding $1 trillion physical AI market,” Dickens said.
Wasson also told EE Times that MIPS was able to leverage its multi-threading architecture for the Atlas portfolio. Analyst firm Tirias Research’s Jim McGregor added, “The MIPS Atlas portfolio aligns tightly with the compute requirements for physical AI and robotics. Working with its customers, MIPS provides a unique multi-threaded architecture with optimized instructions to deliver event-driven low-latency performance along with system-level workload analysis via the Atlas Explorer. MIPS is in a unique position to fill the need for emerging physical AI platforms.”
Wasson said select customers will be able to evaluate the MIPS M8500 real-time compute subsystem with Atlas Explorer in mid-2025, with evaluation boards in fourth quarter 2025, and reference silicon platforms available in 1H26. A MIPS customer automotive platform featuring M8500 is expected to start production in 2027.
From EETimes