The open-source project AxonOS has been unveiled, aiming to create a software infrastructure tailored for systems equipped with a neurocomputer interface. This interface is designed to establish a deterministic execution layer bridging neural hardware and artificial intelligence. The AxonOS kernel is being developed in Rust language, emphasizing predictable real-time execution, a formalized system call ABI, as well as a capability access model devoid of IPC locks and execution time control. The source code for AxonOS is openly available under the MIT or Apache 2.0 licenses.
The project encompasses the development of the kernel/runtime, integration with brain-computer interfaces (BCI), architectural specifications, a consent/permissions model, and real-time analysis tools. Currently, AxonOS is specifically tailored for ARM Cortex-M processors and is undergoing hardware testing and verification of real-time characteristics independently. The architecture of AxonOS views brain-computer interfaces as a multi-level system comprising the brain-computer device, hardware abstraction layer (HAL), deterministic runtime, inter-process communication (IPC), permissions/consent access model, and application/AI layers.
The core concept behind AxonOS is to establish a controlled layer for application execution situated between the brain-computer device and AI. This layer incorporates requirements for determinism, isolation, neural data access, and response time as integral components of the system architecture, rather than relying on the application itself. Another aspect of the project involves analyzing the maximum request execution time, known as Worst-Case Response Time (WCRT), for tasks with fixed priorities. To achieve this, an independent tool named “dy-wcet” is being developed as a “no_std” Rust crate package, built without the use of floating-point operations.