Continuing advances in AI technologies are driving a rapid expansion of AI applications into areas such as smart factories and smart cities. Coupled with the widespread popularity of 4G and 5G communication, GPU performance boosts, and increasing data processing speeds, these developments are spurring burgeoning edge AI market growth. According to GII, the global edge AI hardware market was already valued at US$2.62 billion and is projected to grow at a CAGR of 19.85% over the next five years to US$7.52 billion.
According to eNeural Chairman Jiun-In Guo, although edge AI applications are rapidly expanding, device developers and system solution providers face the challenge of high costs. Without small-scale and high-precision AI models, edge devices need to have GPUs with sufficient computing power. However, not only do GPUs come with high price tags in the range of hundreds of U.S. dollars but they also have high power consumption and thermal management challenges, making the miniaturization of edge devices very difficult. This leads to increased development costs and more importantly, it curbs the growth of edge AI applications. As such, having small-scale and high-precision AI models is critical to developers who strive to keep edge AI device cost and footprint under control.
The market already has a slew of neural-network processing units (NPU) supporting AI tasks, for example, Sunplus C3V, iCatch V37/V57, Ambarella CV2, Realtek AMB82 MINI (RTL8735B), NXP IMX8M+ and Kneron KL series. In view of the market trend, eNeural is dedicated to building small and precise AI models to support customer needs. Its solutions have been adopted by multiple strategic partners.
eNeural builds AI models specifically for NPU platforms, offering unique customization services
Founded in March 2022, eNeural was a spin-off of Intelligent Vision System Lab, National Yang-Ming Chiao Tung University (NYCU). The team has more than a decade of experience in AI vision. Its independently developed AI modeling tool, AI-Craft, accelerates the time-to-market of AI models, bringing AI application products to market six times faster while boosting the precision level. Incorporating fast labeling and model compression techniques, AI-Craft enables partners to develop edge AI devices featuring a lightweight, small footprint, low power consumption, high performance, and high precision.
Guo emphasized that AI models running on NPU platforms on the market are generally developed through open-source projects. They are available in standard versions or manually scaled-down solutions with limited optimization for NPUs. AI-Craft, on the other hand, enables customizable optimizations for various NPUs. Not only can AI-Craft maintain a certain level of recognition precision, but it also makes the AI model up to 90% smaller (depending on the application and requirement). Furthermore, in the case that the precision drops after the AI model is slimmed down, AI-Craft can be used for re-training to raise the precision. This allows eNeural partners to offer a variety of hardware devices suited to different market needs based on only one AI model without the hassle of AI model training and subsequent optimization and scale-down efforts.
eNeural participation at CES caught market attention, successfully expanding AI model applications
In collaboration with strategic partners, eNeural's AI-Craft is being used in advanced driver assistance systems (ADAS), mobile robotics, and AIoT. A leading Taiwan-based manufacturer integrated sensors and cameras on top of AI-Craft, which is the smart core, and introduced QBOX, a vision-based commercial vehicle assistance solution that combines a lane departure warning system (LDWS), forward collision warning system (FCWS), blind spot detection system and blind spot information system (BSD & BSIS) and driver monitoring system (DMS). Apart from this, eNeural is also in talks with potential partners in the healthcare and AIoT sectors while planning to further expand the application scope of AI-Craft.
Guo pointed out that the establishment of eNeural was a result of a program initiated by former Minister of Science and Technology Liang-gee Chen to create four AI innovation and research centers. Under the guidance of the Center of Industry Accelerator and Patent Strategy (IAPS), NYCU, and Taiwan Tech Arena (TTA), eNeural participated at CES 2019 and CES 2023, which boosted the startup's visibility. Taiwan's startup ecosystem has made great strides. With the wealth of opportunities and resources made available by incubators, startups today grow at a faster pace.
Aside from AI-Craft, eNeural has obtained NPU certifications. Going forward, it is able to provide hardware-software integrated AI solutions suited to customer needs, helping them capture tremendous edge AI market opportunities.
The eNeural team
Photo: eNeural
eNeural's AI-Craft for ADAS and smart cockpit applications
Photo: eNeural
