Access Official Technical PDF

Share Summary:LinkedInEmail

Bringing machine learning into a battery-powered device means balancing useful sensing and processing against tight limits on power and available memory. Himax Technologies’ WiseEye WE-I Plus ASIC platform, announced in 2019, was designed to address that challenge for AIoT applications such as smart appliances and surveillance systems. Its approach to ultra-low-power edge AI combines a programmable processor, dedicated hardware acceleration, operating modes for different power states, and a development kit for evaluating applications.

What the WiseEye WE-I Plus platform is designed to do

The WE-I Plus is an AI accelerator-embedded ASIC platform for developers building applications that use convolutional neural network (CNN)-based machine learning. Rather than sending every captured signal to a remote system for analysis, the platform is intended to support processing of data from image, voice and ambient sensors within an AIoT device. This can be relevant when a design has limited on-device memory or must operate within a strict power budget.

Himax describes application areas including home appliances, door locks, televisions, notebooks, building controls and security. These use cases differ in their sensing requirements, but share a need to interpret inputs such as images, audio or environmental signals. The platform is also presented as suitable for audio, video and signal-processing applications developed with popular machine-learning frameworks.

Hardware acceleration and programmable processing

At the core of the WE-I Plus is a programmable processor with enhanced digital signal processing (DSP) features. It is paired with a power-efficient hardware accelerator for CDM, HOG and JPEG processing. The announcement identifies real-time motion detection, object detection and image processing as target tasks for this combination.

For system designers, the distinction between programmable processing and dedicated acceleration matters. A programmable processor provides a software execution environment, while specialized hardware can handle supported processing tasks. The platform brings both elements together, although the specific division of work will depend on the application and its implementation. Developers need to assess their model, sensor inputs and processing needs against the available software libraries and hardware functions.

The platform’s relevance is not limited to visual sensing. Himax describes a software library and tool chain for implementing machine learning on captured data from image, voice and ambient sensors. The supplied DSP library complements the machine-learning library, giving developers resources for both ML and signal-processing tasks.

Power management for sensing and inference

Low-power design is addressed at both the ASIC and power-management levels. The WE-I Plus includes an embedded low-dropout regulator (LDO) and a multi-state power management unit (PMU). According to Himax, the design supports shutdown, always-on sensing (AoS), and CV operation modes intended for efficient operation.

These modes reflect an important consideration in battery-powered sensing systems: a device may need to monitor its environment without keeping every part of its processing system active continuously. Always-on sensing is intended to support that kind of operating pattern, while shutdown and other operating states provide options for managing power. The appropriate mode depends on how an application is designed to sense and respond; the announcement does not specify power figures or battery-life results.

Security features for connected AIoT devices

AIoT deployments also need to account for security across device startup, software updates and the handling of metadata. Himax says the WE-I Plus integrates hardware and software security features that include secure boot, secure over-the-air (OTA) updates, and security metadata output over Transport Layer Security (TLS).

These capabilities are relevant to device-level security planning, but they do not replace the broader work of designing and managing a secure system. Product teams still need to consider how the ASIC’s features fit with their firmware, communications, update processes and application requirements.

Evaluation kit supports early development

The WE-I Plus evaluation kit (EVK) is intended to help application developers assess the platform and advance system design. Its listed components include the WE-I Plus ASIC, an AoS sensor, a passive infrared (PIR) sensor, a temperature sensor, an ambient light sensor, microphones and LEDs. Wi-Fi and USB extension boards are also provided, along with UART, I2C and SPI links to a PC.

On the software side, the EVK includes machine-learning and DSP libraries and supports the use of an in-circuit emulator (ICE) in a programming environment described as user-friendly. For integrators, this combination offers a starting point for exploring how sensor inputs, processing software and hardware interfaces work together before carrying a design further into product development.

Practical considerations for engineering teams

When evaluating an embedded AI platform such as the WE-I Plus, engineering teams can use the announced feature set to frame several design questions:

  • Which sensor inputs and processing tasks are central to the application?
  • Can the model and its supporting software fit within the device’s memory and power constraints?
  • How should sensing and processing be distributed across the platform’s operating modes?
  • How will secure boot, OTA updates and TLS-protected metadata fit into the product’s security design?
  • Can the EVK’s sensors, interfaces and software libraries support an effective early evaluation?

The answers will depend on the product’s requirements and implementation. The announcement provides no comparative performance figures, power measurements or model-capacity specifications, so these should be established through application-specific evaluation rather than assumed.

Conclusion

The WiseEye WE-I Plus illustrates a platform-level approach to ultra-low-power edge AI: combine programmable processing and dedicated acceleration with power-management options, security features and development tools. For automation and AIoT teams, its practical value lies in evaluating whether that combination fits a particular sensing task, operating profile and security architecture. The EVK and associated software libraries provide a route to explore those requirements during system design, while application-specific testing remains essential to confirm suitability.

Access Official Technical PDF

About The Author