# ACAI: Protecting Accelerator Execution with Arm Confidential Computing Architecture

**Authors:** Supraja Sridhara, Andrin Bertschi, Benedict Schlüter, Mark Kuhne, Fabio Aliberti, Shweta Shinde  
**arXiv:** [2305.15986](https://arxiv.org/abs/2305.15986)  
**Date:** 2023-05-25  
**Summary page:** https://www.summarizepaper.com/en/paper/2305.15986v2/

## Key points

- Authors address limitations of existing trusted execution environments in CPUs for secure usage of accelerators like GPUs and FPGAs
- Focus on reevaluating Arm Confidential Computing Architecture (CCA) design within Armv9-A to bridge the gap effectively
- CCA enables confidential VMs to utilize accelerators as a primary abstraction
- Introduce ACAI, a solution based on CCA principles, to extend security invariants for device-side access
- Experimental evaluations on GPU and FPGA platforms demonstrate feasibility of ACAI while maintaining stringent security guarantees
- Research contributes insights towards advancing confidential computing paradigms and enhancing security measures in modern computing architectures

## Summary

In their paper titled "ACAI: Protecting Accelerator Execution with Arm Confidential Computing Architecture," authors Supraja Sridhara, Andrin Bertschi, Benedict Schlüter, Mark Kuhne, Fabio Aliberti, and Shweta Shinde address the limitations of existing trusted execution environments in CPUs regarding the secure usage of accelerators like GPUs and FPGAs. They focus on reevaluating the Arm Confidential Computing Architecture (CCA) design within Armv9-A to bridge this gap effectively. The authors highlight that CCA presents suitable abstractions and mechanisms that enable confidential virtual machines (VMs) to utilize accelerators as a primary abstraction. To enhance this capability further, they introduce ACAI, a solution based on CCA principles. ACAI extends CCA security invariants to facilitate device-side access, thereby addressing critical security gaps that hinder the secure utilization of accelerators by confidential VMs. Through experimental evaluations conducted on GPU and FPGA platforms, the authors demonstrate the feasibility of ACAI while upholding stringent security guarantees. Their findings underscore the potential of leveraging CCA-based solutions like ACAI to enhance confidential computing capabilities in CPUs and enable secure accelerator execution within trusted execution environments. This research contributes valuable insights towards advancing confidential computing paradigms and ensuring robust security measures in modern computing architectures.