Developing Edge AI Applications with ADI's MAX78002 MCU

The financial arms race among tech giants to commercialize generative artificial intelligence (GenAI) somewhat obscures other important AI efforts, notably at the edge of the network where vendors are eager for AI applications able to run on IoT devices that are typically constrained by limited memory, bandwidth, and power.

A microcontroller unit (MCU) from Analog Devices, Inc., can address edge processing limitations with an integrated low-power convolutional neural network (CNN) accelerator for processing AI inferences on battery-powered devices.

Whereas GenAI investments are largely oriented around accumulating vast amounts of data and processing capabilities requiring large-scale data centers and tremendous amounts of energy, edge AI revolves around efficiently running data locally through models that can identify objects, analyze medical images, and process automobile camera feeds to recognize obstacles, pedestrians, and road signage for safe driving applications.

CNNs can process image data at the edge, detecting anomalies and monitoring the health of equipment on the factory floor. They can also detect pests and crop health in agricultural settings and process images from drones, robots, and smart cameras.

Optimized for deep CNNs

ADI's MAX78002 is an ultra-low-power, advanced system-on-chip that features an Arm Cortex-M4 processor with floating point unit (FPU) and a hardware-based accelerator optimized for deep CNNs and tasks requiring object recognition.

Weights—or parameters—interconnect the neurons in a neural network to control its behavior. ADI's CNN engine has a weight storage memory of 2 MB that can support 1, 2, 4, and 8-bit weights, as well as complex neural network models up to 16 million weights. This enables advanced AI applications on edge devices, and because the CNN weight memory is SRAM-based, model updates can be made on the fly.

The CNN accelerator provides a programmable input image size up to 2048 pixels x 2048 pixels, providing developers with the flexibility to design applications that can handle high-resolution medical imaging or process smaller input sizes on resource-constrained devices.

Programmable network depth up to 128 layers makes it possible to balance expressiveness and efficiency of applications. Additionally, programmable per-layer network channel widths up to 1024 channels provide the ability to utilize wider channels for capturing richer features, or use narrower widths that save memory and computational resources.

The MAX78002 supports multiple high-speed and low-power communications interfaces, including I²S, MIPI CSI-2 serial camera, parallel camera (PCIF), and SD 3.0/SDIO 3.0/eMMC 4.51 secure digital. This makes it highly suited to a wide range of AI applications, including industrial sensors, process control, inline quality assurance vision systems, portable medical diagnostics equipment, factory robots, and drone navigation.

Power management is a crucial issue

Ultra-low-power microcontrollers are critical to edge AI applications, particularly so when reliant on battery-powered IoT devices. ADI says the MAX78002 consumes just microjoules of energy when processing AI inferences.

This AI MCU boasts a built-in, single-inductor multiple-output (SIMO) switch-mode power supply (SMPS) that supports a supply voltage range of 2.85 V to 3.6 V to accommodate diverse power sources. Additionally, it enables optional control of external switches to provide CNNs with dedicated power from external sources.

The MAX78002's power management unit (PMU) features intelligent, precision control of power distribution to CPUs and peripheral circuitry to support high-performance operation with minimized power consumption.

A monolithic power supply architecture enables operation from a single lithium cell. Three buck regulator outputs in the SIMO are voltage programmable to ensure optimal power consumption efficiency. Vendors can reduce the bill of materials for circuit designs as the MAX78002 needs just a single inductor/capacitor.

An integrated dynamic voltage scaling (DVS) controller can adaptively adjust voltage to realize a reduction in dynamic power consumption. Using a fixed high-speed oscillator and V_COREA supply voltage, the DVS controller can operate the Arm core at the lowest practical voltage, providing product designers the ability to balance performance requirements with power constraints. An Arm peripheral bus interface provides control and status access.

Large on-chip system memory of 2.5 MB flash for the microcontroller core ensures nonvolatile storage of program and data memory, and internal 384 KB SRAM provides low-power retention of application information in all power modes except POWER DOWN.

Facilitating MAX78002 applications

ADI offers the MAX78002EVKIT (Figure 1) evaluation kit that provides valuable resources for building AI applications with the MCU, including a 2.4-inch TFT display that enhances interactive UI development and visualization of the results of AI inference processes.

Figure 1: The evaluation kit for MAX78002 applications includes a 2.4-inch TFT display for interactive UI development and a secondary display for tracking power consumption. (Image source: Analog Devices, Inc.)

With the evaluation board, energy consumption of the MAX78002 is tracked by a power accumulator, with formatted results presented on a secondary TFT display.

The evaluation kit includes USB 2.0, SWD JTAG headers, UART access over USB, and dual industry-standard QWIIC connectors, facilitating debugging, programming, and interfacing with other devices.

Conclusion

The limited memory, bandwidth, and power consumption of edge IoT devices is a major challenge in developing AI edge applications. ADI's MAX78002 MCU provides a clear path to develop a broad range of power-efficient AI applications with inference capabilities. With the MAX78002EVKIT, developers have ready access to tools for rapid prototyping, touch-enabled UI development, peripheral integration, and energy consumption tracking. Check out an unboxing video for a quick overview of what’s included in the evaluation kit.