Courtesy: NXP Semiconductors
AI integration into wearable technology is experiencing explosive growth and covering a variety of application scenarios from portable assistants to health management. Their convenience of operation has also become a highlight of AI glasses. Users can easily access teleprompting, object recognition, real-time translation, navigation, health monitoring, and other operations without physically interacting with their mobile phones. AI glasses offer a plethora of use cases seamlessly integrating the digital and real worlds, powering the next emerging market.
The Power Challenge: Performance vs. Leakage
The main challenge for AI glasses is battery life. Limited by the weight and size of the device itself, AI glasses are usually equipped with a battery capacity of only 150~300mAh. To support diverse application scenarios, related high-performance application processors mostly use advanced process nodes of 6nm and below. Although the chip under this process has excellent dynamic running performance, it also brings serious leakage challenges. As the process nodes shrink, the leakage current of the silicon can increase by an order of magnitude. The contradiction between high leakage current and limited battery capacity significantly reduces the actual usage time of the product and negatively affects the user experience.
The chip architect is forced to weigh the benefits of the various process nodes, keeping in mind active power as well as leakage. With the challenge of minimising energy usage, many designs have taken advantage of a dual chip architecture, allowing for lower active power consumption by using the advanced process nodes, while achieving standby times with much lower leakage through the more established process nodes.
Solving the Power Problem: Two Mainstream Architectures
Currently, AI glasses solutions on the market mainly use two mainstream architectures:
“Application Processor + Coprocessor” Architecture
The “application processor + coprocessor” solution can bring users the richest functional experience and maximise battery life. The application processors used in AI Glasses are based on advanced processes, focusing on high performance, usually supporting high-resolution cameras, video encoding, high-performance neural network processing, and Wi-Fi/Bluetooth connectivity. In turn, coprocessors steer towards mature process technologies, focusing on lower frequencies to reduce operating and quiescent power consumption. The combination of lower active and standby power enables always-on features such as microphone beam forming and noise reduction for voice wake-up, voice calls, and music playback.
“MCU-only” Architecture
The “MCU-only” solution opens the door to designs with longer battery life, lighter and smaller frames, giving OEMs an easier path towards user comfort. With weight being one of the most important factors in the user experience of glasses, the MCU-only architecture reduces the number of components as well as the size of the battery. The weight of the glasses can be brought down to within 30g.
The strategy of an MCU-only architecture puts more emphasis on the microcontroller’s features and capabilities. Many features of the AP-Coprocessor design are expected within the MCU design. It is therefore critical to include features such as NPU, DSP, and a high-performing CPU core.
NXP’s Solution: The i.MX RT Family as the Ideal Coprocessor
The i.MX RT500, i.MX RT600 and i.MX RT700 has three chips in NXP’s i. MX RT low-power product family. These chips, as coprocessors, are currently widely used in the latest AI eyewear designs for many customers around the world. The i.MX RT500 Fusion F1 DSP can support voice wake-up, music playback, and call functions of smart glasses. The i.MX RT600 is mainly used as an audio coprocessor for smart glasses, supporting most noise reduction, beamforming, and wake-up algorithms. The i.MX RT700 features dual DSP (HiFi4/HiFi1) architecture and supports algorithmic processing of multiple complexities, while enabling greater power savings with the separation of power/clock domains between compute and sense subsystems.
How the i.MX RT700 Maximises Battery Life
As a coprocessor in AI glasses, the i.MX RT700 can flexibly configure power management and clock domains to switch roles based on different application scenarios: it can be used as an AI computing unit for high-performance multimedia data processing, and it can also be used as a voice input sensor hub for data processing in ultra-low power consumption.
AI glasses mainly rely on voice control to achieve user interaction, so voice wake-up is the most commonly used scenario and the key to determining the battery life of AI glasses. In mainstream use cases, the coprocessor remains in active mode at the lowest possible core voltage levels, awaiting the user’s voice commands, quickly switching to speech recognition mode with noise reduction in potentially noisy environments. Based on this user scenario, the i.MX RT700 can be configured to operate in sensor mode; at this time, only a few modules, such as HiFi1 DSP, DMA, MICFIL, SRAM, and power control (PMC), are active. The Digital Audio Interface (MICFIL) allows microphone signal acquisition; DMA is used for microphone signal handling; HiFi1 is used for noise reduction and wake-up algorithm execution, while the compute domain is in a power-down state.
Other low-power technologies included in the RT700, such as distortion-free audio clock source FRO, microphone module FIFO, and hardware voice detection (Hardware VAD), DMA wake-up also ensures that the system power consumption of i.MX RT700 voice wake-up scene can be under 2 mW, maximising power consumption while continuously monitoring.
RT700 also powers MCU-only
For display-related user scenarios, the i.MX RT700 can be configured in “High Performance Mode”, where the Vector Graphics Accelerator (2.5D GPU), Display Controller (LCDIF), and Display Bus (MIPI DSI) are enabled. While enabling high performance, the compute domain also takes advantage of low-power technologies such as MIPI ULPS (Ultra Low Power State), dynamic voltage regulation within the Process Voltage Temperature (PVT) tuning, and other low-power technologies.
With the continuous integration of intelligent hardware and artificial intelligence, choosing the right low-power high-performance chip has become the key to product innovation. With its deep technology accumulation, the i.MX RT series provides a solid foundation for cutting-edge applications such as AI glasses.
