Skip to content
FUTURE RESEARCH

Future Research Directions

Long-term research goals and moonshot projects for the EmbeddedOS ecosystem. These are the problems we are working on for the next 5–10 years.

Formal Verification

2026–2028

Apply TLA+ and Coq formal verification to the EoS kernel scheduler, EIPC protocol, and secure boot chain. Goal: mathematical proof of absence of deadlocks, race conditions, and memory safety violations in the kernel core.

Formalize EIPC message ordering invariants in TLA+
Verify MPU configuration correctness with Coq
Automated verification in CI pipeline for kernel patches

Closed-Loop Neural Interfaces

2026–2029

Build the first closed-loop BCI system on EoS: real-time neural decoding → stimulation feedback loop with <5ms round-trip latency. Targeting motor cortex prosthetics and seizure detection/suppression.

1024-channel spike sorting at 30 kHz on RK3588S NPU
Closed-loop stimulation with <5ms decode-to-stimulate latency
FDA Breakthrough Device designation pathway

Distributed Embedded OS

2027–2030

Extend EoS to support distributed operation across multiple physical devices as a single logical system. A swarm of 100 microcontrollers appears as one EoS instance with distributed task scheduling and shared memory.

Distributed EIPC over mesh networking (Thread/Zigbee)
Consensus-based distributed task scheduler
Single-image OTA update for device swarms

Space-Grade EoS

2027–2031

Harden EoS for space environments: radiation-tolerant memory management, triple-modular redundancy for critical tasks, and DO-178C Level A certification. Targeting CubeSat and small satellite platforms.

Radiation-hardened memory allocator with ECC scrubbing
TMR task execution with automatic fault recovery
DO-178C Level A certification for EoS kernel

Sub-mW AI Inference

2026–2028

Achieve <1mW AI inference for always-on keyword spotting, gesture recognition, and anomaly detection on Cortex-M0+ class devices. Combining extreme quantization (2-bit), sparse activation, and custom SIMD kernels.

2-bit quantized keyword spotting model at 800μW
Sparse activation runtime for Cortex-M0+
EAI manifest support for ultra-low-power model variants

Neuromorphic Computing

2028–2032

Integrate neuromorphic processors (Intel Loihi 2, BrainScaleS) into the EoS HAL. Enable spiking neural network inference at 1000x lower energy than conventional deep learning for always-on sensor processing.

EoS HAL driver for Intel Loihi 2 via USB3
SNN inference runtime with EIPC integration
Benchmark: SNN vs CNN on EAI edge workloads