Secure Monitor Calls (SMCs) for AI-Centric Embedded Platforms: Bridging Software Flexibility and Hardware Control

Abstract

Secure Monitor Calls (SMCs) are critical for AI-focused embedded platforms, acting as the interface between non-secure operating systems and secure firmware. ARM TrustZone enforces hardware isolation, allowing secure interaction between ARM cores, AI accelerators, and programmable logic. Modern embedded AI systems require more than processor-level control; they need advanced power, thermal, and resource management with deterministic behavior for safety. Two widely used interfaces rely on Secure Monitor Calls (SMCs): the ARM-standard PSCI for coordinating power states, and the Xilinx-AMD-specific EEMI for managing voltage, frequency scaling, and thermal limits. SMCs rely on register-based parameter passing to ensure secure, cross-platform communication. While platforms may add extensions for hardware differences, this model preserves portability and simplifies integration. Compared to other methods, it improves security, determinism, and system coordination. As systems evolve toward adaptive computing, SMC implementations must expand to support new AI workloads while remaining compatible with existing frameworks.