Slide 1: RF Modulation Classification at the Edge
RF Modulation Classification at the Edge Communication Reference Platform July 2026 © 2026 BrainChip Holdings Ltd. All rights reserved.
Slide 2: Presenters: Dr. Steven D. Harbour and Aras Pirbadian
Placeholder for Bio slide Dr. Steven D. Harbour Aras Pirbadian Technical Fellow Sr. Machine Learning Engineer BrainChip BrainChip © 2026 BrainChip Holdings Ltd. All rights reserved. 2
Slide 3: Who is BrainChip? The Pioneers of Neuromorphic AI
Who is BrainChip? The Pioneers of Neuromorphic AI Who We Are What is Neuromorphic AI The world’s first commercial producer of “Neuro” = brain. “Morphic” = shaped like. neuromorphic AI processors Literally: AI shaped like a brain Publicly listed deep-tech company (ASX: BRN) Mimics how neurons fire on
Slide 4: Why Akida Does More with Less
Why Akida Does More with Less Traditional AI Chips Brute-Force Every Calculation. Akida Only Computes What Matters — Just Like Your Brain. Conventional NPU Akida NPU Compute all data Compute only event data "brute force" vs. "elegant, like the brain" What’s the Result? 10X Computational Efficiency m
Slide 5: Powering Next-Gen Physical AI
BrainChip Offerings Powering Next-Gen Reference Physical AI Platforms Dev Kits & Modules On-Device Physical AI, Near the Sensor Akida Brain-Inspired Neuromorphic Computing for Models Ultra-Low Power Event-Based Processing Akida Scalable Platform from IP to Solutions Tools Akida Silicon License IP ©
Slide 6: Akida Communication Reference Platform
BrainChip Offerings AkidaTM Communication Reference Platform Identify and classify modulations, in real time, at the tactical edge, on milliwatts of power. © 2026 BrainChip Holdings Ltd. All rights reserved. 6
Slide 7: The Edge Signal Intelligence Challenge
The Challenge The Edge Signal Intelligence Challenge Mission-Critical SWaP-C Legacy DSP Limits Cloud Dependency Risks Defense platforms require minimal size, Traditional DSP chains cannot adapt to Off-board analysis adds unacceptable weight, power, and cost—ruling out high- complex, evolving threats
Slide 8: Why Akida for Signal Intelligence
Mission-Ready Advantages Why Akida for Signal Intelligence Low-SWaP Intelligence 100% Data Sovereignty Ultra-Low Latency Advanced classification in All processing stays on-device Millisecond-level reaction to battery-powered devices with — a zero-emission profile jamming and threats, with no no ther
Slide 9: A Complete RF-to-Decision Pipeline
System Overview A Complete RF-to-Decision Pipeline The reference design pairs a software-defined radio with a Raspberry Pi 5 host and the Akida AKD1000 accelerator on an M.2 module — a full modulation classification pipeline in a palm-sized footprint. AI Processor SDR Front End AKD1000 70 MHz – 6 GH
Slide 10: Built on the Epiq Sidekiq Stretch
SDR Benefits Built on the Epiq Sidekiq Stretch The reference platform pairs Akida with the Epiq Sidekiq Stretch — a compact, low-power M.2 software-defined radio purpose-built for size-, weight-, and power-constrained missions. Compact M.2 Form Factor Mission-Grade Power Efficiency A full RF transce
Slide 11: Lab-Validated Performance
Proven Results Lab-Validated Performance Model size: 3.7M MACs and 104k parameters 90%+ <1 W Classification Power at continuous accuracy inference 0.11 W 0.26 ms Idle power Inference latency (measured by RTX) per frame 2,734 23 Frames per second, Modulation continuous schemes covered Confusion matri
Slide 12: Built for the Missions That Matter
Target Applications Built for the Missions That Matter Electronic Warfare UAV / Drone SIGINT Real-time threat detection and jammer classification at Passive signals intelligence on small unmanned the point of signal capture — millisecond response with platforms, with dramatically reduced data-link n
Slide 13: RF Modulation Classification in Action
Demos RF Modulation Classification in Action Insert Kylie’s Video © 2026 BrainChip Holdings Ltd. All rights reserved. 13
Slide 14: Key Takeaways
Key Takeaways We Are…. Showcasing RF modulation classification implemented with neuromorphic deep learning Introducing ultra-low SWaP-C signal intelligence algorithms Delivering an end-to-end platform to show it is real! The platform allows: Recording your own RF dataset, training and validating you
Slide 15: Q & A
Q&A © 2026 BrainChip Holdings Ltd. All rights reserved. 15