MIPS will use Apex Design Optimization in conjunction with its workload-based development approach and software.
According to Xcelsa, Apex combines two components called the Verified Intelligence Stack and the Physical Intelligence Stack.
The former is designed to verify, through formal equivalence, that the modifications made during optimization maintain the functional behavior of the reference design.
The latter analyzes the physical consequences of the proposed modifications. Xcelsa notes that this capability can reason about these effects without relying on EDA tools during the process.
The platform can explore thousands of design alternatives, according to the companies, with the goal of optimizing power, performance, and area parameters (PPA).
MIPS and Xcelsa present a specific result obtained during work on a production-scale MIPS design.
In this test, Apex addressed a route with a significant timing violation in a load-store unit.
According to the companies, the platform resolved the violation in less than three hours and achieved a 33% reduction in the critical path delay.
The optimization involved rewriting a substantial portion of the design block. Subsequently, the result underwent formal verification, which determined its functional equivalence with the reference design. This 33% reduction corresponds to that specific route and that particular case.
MIPS estimates that performing comparable work manually would have required approximately one month of an engineer's time.
One of the platform's key features is the combination of automated design modification and formal equivalence verification.
In the provided example, Apex modified a circuit block to resolve the timing issue, and the resulting design was then formally verified to retain the functional behavior of the original.
This allows us to separate two different objectives: improving certain implementation parameters and verifying that the transformations made have not altered the functionality that is intended to be preserved.
MIPS frames this collaboration within its strategy of platforms geared towards specific workloads.
The approach involves linking software requirements with hardware architecture and implementation decisions. In applications such as automotive, robotics, or industrial control, parameters such as deterministic performance, energy consumption, real-time response, and functional safety requirements may be involved.
The addition of Apex aims to expand the number of alternatives that can be explored during design and optimization iterations.
AI, automotive, and robotics applications:
MIPS and Xcelsa are directing their collaboration toward the development of custom silicon for AI, automotive, aerospace, industrial, robotics, communications infrastructure, and other embedded systems.
MIPS also links the agreement to its platform strategy for Physical AI, a term it uses for artificial intelligence systems that interact with physical environments.
The alliance expands the MIPS ecosystem with Xcelsa's optimization capabilities, but the announcement does not include any new processors, commercial ASICs, or jointly developed physical devices.
