Massive growth in global connected devices is also driven by edge AI capabilities Two key technologies are underpinning edge AI: Low-power RF and low-power AI compute HaiLa and BrainChip partner to show both elements today
www.linkedin.com
View attachment 88347
Article by Patricia Bower, HaiLa VP Product What’s the ‘Edge’ in Edge AI? The amount of data generated by billions of wireless connected devices—such as sensors, wearables, and smart appliances—is growing at an incredible rate. Analysts forecast global growth from 20 billion devices today to over 40
www.linkedin.com
Cloud based AI vs Edge AI Topology
Connected, Low-power, Intelligent Devices at the Edge
4,385 followers
July 10, 2025
Article by
Patricia Bower,
HaiLa VP Product
What’s the ‘Edge’ in Edge AI?
The amount of data generated by billions of wireless connected devices—such as sensors, wearables, and smart appliances—is growing at an incredible rate. Analysts forecast global growth from 20 billion devices today to over 40 billion by 2030 (IoT Analytics, State of IoT 2024).
Using today’s communications networks, user data is sent from (for example) the smartwatch on your wrist to datacenters located miles from you or even across a continent. The data is processed in these facilities by clusters of artificial intelligence (AI) or machine learning (ML) systems and is generally referred to as cloud compute. There are three issues with this model:
- The time it takes for the data to traverse communications networks to the datacenter (latency)
- The heavy load on datacenter AI clusters, and resulting power consumption, for processing raw data from billions of devices
- The higher potential for personal data privacy breaches
The growing need for faster responses (real-time decision making), reduced loading on datacenters (cloud offload), and improved privacy (anonymized data) has therefore driven the rise of edge AI—where intelligent processing happens directly on the device itself, such as in a sensor or wearable. At the core of this shift are neural processing units (NPUs)—specialized AI processors that enable devices to "think locally” and to rely on optimized AI and ML training models.
Chip technology for datacenter AI processing is extremely energy intense. In these AI clusters, where complex computational tasks - such as climate modelling - are performed, compute operations are measured in exaFLOPS, or a million, trillion floating-point operations per second. Edge AI NPUs can support up to a few teraOPS, or trillion operations per second OPS, and are purpose-built to be extremely power efficient. This is important as many connected devices run on batteries.
What Applications are Driving Edge AI?
Edge AI is already transforming many industries. Here are some notable applications:
Patient Monitoring
Hospitals and home-care settings are increasingly using AI-powered sensors to continuously monitor vital signs such as heart rate, respiration, and oxygen levels. By analyzing these signals locally, devices can detect early warning signs of health deterioration and alert caregivers—without the need to stream personal health data to the cloud. This approach protects patient privacy and saves bandwidth.
Personal Health Monitoring
Wearable devices like fitness trackers and smartwatches are incorporating edge AI to offer more intelligent insights into sleep patterns, stress levels, and activity trends. These devices process sensor data in real time, providing instant feedback and recommendations on the go while preserving battery life.
Retail
In retail environments, smart cameras and sensors are used for foot traffic analysis, shelf inventory monitoring, and customer behavior tracking. Edge AI allows these insights to be processed locally without storing or transmitting images, which supports both privacy and efficiency. For example, a store can detect when a shelf is empty and trigger restocking alerts automatically. Using a battery powered, modular solution with integrated, low-power Wi-Fi provides a simple installation into legacy environments.
Industrial Asset Monitoring and Anomaly Detection
Factories and industrial sites rely on sensors to monitor machinery. Edge AI enables these sensors to learn the normal operating patterns of equipment and detect anomalies like unusual vibrations or temperature spikes—early signs of failure. This allows for predictive maintenance, reducing downtime and maintenance costs.
The ability to collect intelligent data directly from edge devices reduces the need for cloud infrastructure and data transmission, which means that edge AI also lowers the overall cost of system deployment and operation.
Leveraging Low-power RF and Neuromorphic AI for Efficient Edge AI
A major technical challenge in deploying edge AI is power consumption. Devices like health monitors or industrial sensors are often required to run for months or years on small batteries—or in some cases, without batteries at all. As mentioned, NPUs optimized for low power are an essential element to perform AI tasks like image recognition or anomaly detection using minimal energy. To address this need,
BrainChip’s Akida™ neuromorphic technology relies on the principle of sparsity for power efficient, event-driven AI compute.
The inferenced data from NPU AI processing is only part of the equation. Devices must also have the ability to connect and collate this intelligent data from multiple devices in a local or personal area network. This is where low-power radio frequency (RF) technologies like Wi-Fi and Bluetooth Low Energy come in and where HaiLa is setting a new paradigm in just how efficient data transmission can be over these protocols. HaiLa’s extreme low power radio communications technology, paired with power-optimized edge compute, allows devices to send and receive inferenced data without draining energy reserves, making them perfect partners for edge AI systems.
HaiLa and BrainChip: Sensors Converge 2025 Connected Edge AI Demo
At Sensors Converge in Santa Clara this year, HaiLa and BrainChip joined forces to demonstrate object classification via extreme low-power Wi-Fi.
HaiLa and BrainChip collaborated to showcase very low power Connected Edge AI Object Classification
This hardware prototype includes a camera module to capture object images, BrainChip’s NPU test platform which is pre-trained to recognize objects from a camera image capture, and HaiLa’s BSC2000 extreme low-power Wi-Fi radio chip which transmits the image class and type data via Wi-Fi to display as a simple icon on a dashboard.
The demo illustrates the potential use of connected, edge AI for applications using image recognition and classification in industrial, retail, and medical sectors where cost effective, low latency, and private data connections are key requirements.
The Future of Edge AI: Connected, Low-power, Intelligent Devices Everywhere
HaiLa’s core specialization in extreme low-power radio technology over standard protocols like Wi-Fi, Bluetooth, and even cellular, delivers one of the critical enablers of pervasive edge AI: extreme low-power data transmission. Together with BrainChip’s innovative edge compute, this opens up a broad range of options for end-users to support energy-efficient, on-device AI.
Edge AI represents a powerful shift in how intelligent systems operate—bringing the power of AI directly to the devices at the heart of the connected world. By combining efficient NPUs with low-power wireless communication, edge AI systems can run independently, securely, and with minimal energy use. As more industries adopt this technology, we can expect smarter, more responsive, and more sustainable solutions for multiple applications.
Contact us to learn more:
info@haila.io
sales@brainchip.com