Embedded Files
Moonlight Technologies Announces "HPDBN" Edge Whole-Brain Architecture for EvoSpikeNet
Yokohama, August 31, 2026 – Moonlight Technologies Inc. (CEO: Masahiro Aoki; Headquarters: Yokohama) has announced "HPDBN" (Heterogeneous Polymorphic Distributed Brain Node), a standard architecture for its distributed neuromorphic framework "EvoSpikeNet." HPDBN enables execution with a clear separation between sensor input and motor/actuator control across diverse hardware environments, including edge devices.
Background In robotics, drones, and industrial edge machinery, embedded hardware is increasingly heterogeneous—spanning CPUs, GPUs, SNNs, and NPUs. However, the responsibilities for sensor processing, cognitive inference, and actuator control often become tightly coupled in code. This tight coupling creates significant challenges for ensuring safety during network disconnections and maintaining code portability across different hardware platforms. EvoSpikeNet's HPDBN addresses this by defining a standardized loop that explicitly decouples cognitive inference from hardware-level control.
Feature Overview
Locally Self-Contained Inference Across Heterogeneous Hardware: Each edge node maintains Local Spatial Nodes (spatial awareness) and a Local PFC (Prefrontal Cortex model for cognitive integration), allowing local inference to continue uninterrupted even if external communication fails.
Separation of Sensor and Control Responsibilities: The standard loop consists of five stages: Sensors → Spatial Nodes → Features (features, confidence, TTC candidates) → PFC → Runtime (CPU/SNN/NPU) → Application safety/control adapter. Raw sensor data is never passed directly to the PFC or external networks; instead, only versioned features and tracking summaries are transmitted. Actuator control and hardware-specific safety mechanisms are strictly decoupled as responsibilities of the consuming application's adapter layer.
Automatic Fallback Upon Disconnection: When network degradation or keepalive expiration is detected, the system halts external collaborative computing and maintains local inference using cached data with defined expiration limits—driven strictly by local Spatial Nodes and the local runtime. Upon connection recovery, stale control plans are rejected unless authentication, monotonic time, plan generation, node ID, and model version are fully verified.
Integration with Device Abstraction: By pairing HPDBN with existing device plugin mechanisms (supporting CPU, GPU, Jetson, Loihi, EdgeTPU, etc.), developers can execute identical cognitive inference logic regardless of the underlying hardware.
Benefits for Developers By adhering to the HPDBN standard, developers can implement cognitive inference logic without coupling it to hardware or application-specific safety constraints. Responsibilities for sensor adapters, control adapters, Safety Barriers, and hardware monitoring are explicitly shifted to the application side. This separation allows developers to combine EvoSpikeNet Core's cognitive inference, peer collaboration, and device abstraction with custom implementations tailored to specific airframe or robot safety certifications (e.g., DO-178C).
Acceptance Criteria The Core implementation must satisfy the following criteria to meet the standard:
Raw sensor data must not be inadvertently transmitted externally.
Local inference must not halt upon network disconnection.
Stoppages in PFC, Spatial Nodes, or runtime must be detectable.
Outputs must include input expiration, quality metrics, model versions, and reason codes.
Worst-Case Execution Time (WCET), temperature, power consumption, and memory usage must be measurable across CPUs, GPUs, and NPUs.
For detailed specifications, please refer to the "HPDBN SDK Integration Guide" on our documentation site. Note that this standard represents Core-side design specifications and does not constitute airborne airworthiness or safety certification in itself.
Executive Statement Masahiro Aoki, CEO of Moonlight Technologies Inc., stated:
"In field environments where diverse edge hardware coexists, explicitly decoupling cognitive inference from hardware control responsibilities is essential to achieving both safety and portability. HPDBN implements this design philosophy as a native standard within EvoSpikeNet."
Company Overview
Company Name: Moonlight Technologies Inc.
CEO: Masahiro Aoki
Established: 2025
Business Activities: AI framework development, distributed brain simulation, and neuromorphic computing
Website: https://www.moonlight-tech.biz
Contact Information
Email: info@moonlight-tech.biz
Page updated
Report abuse