By Stephen Las Marias, EE Times Asia
TAINAN, Jan. 13, 2024 /PRNewswire/ — The proliferation of AI-driven applications across diverse sectors has sparked a wave of AI progress in 2023, spanning data centers, edge infrastructure, and endpoint devices. This trend also signifies a significant revenue opportunity for the semiconductor industry for the year.
Himax: Enabling AI Everywhere
Gartner Inc. forecasts that AI chips will contribute $53.4 billion in revenue to the semiconductor industry in 2023, signifying a significant 21% surge from 2022. The market analysis firm also foresees continued double-digit revenue growth to $67.1 billion in 2024. The trajectory is expected to soar, reaching an impressive $119.4 billion by 2027.
Notwithstanding these positive forecasts, the AI semiconductor sector encounters persistent challenges, particularly in performance and power efficiency. Further efforts are needed to enhance and refine designs that enable the low power execution of AI-based workloads.
On top of this, everybody wants their AI applications, particularly for endpoint devices, to include user inference, says Mark Chen, Vice President of Smart Sensing business at Himax Technologies Inc.
"Himax is a fabless IC design firm in Taiwan focusing on solutions including display drivers, image sensors, and AI processors. Himax is also known as the global market share leader in automotive display market with dominated 40% market share." says Chen, who leads the image sensor and AI processor business units of the company.
"Many believe Machine Learning requires extensive input data and power-intensive computing to train models for intricate image recognition. However, in the era of IoT, everything is gradually shifting to the extreme edge." Said Mr. Chen. "What strategies can a semiconductor companies employ to bring AI into every realm, particularly in the consumer product segment? Himax claims that, through creating solutions that are ultralow power and more cost-effective."
"This is why we continue to expand our AI roadmap and roll out WiseEye2, built upon the success of its predecessor chip, WE1." states Chen. "WiseEye2 stands as an ultralow power AI processor specifically designed for on-device computer vision tasks. It boasts 60mW facial mesh processing, a dedicated microNPU core, advanced power management, ample on-chip memory, as well as compatibility with an open software platform."
WiseEye2 caters to applications that necessitate immediate responsiveness and quick decision-making, such as industrial quality inspection, predictive maintenance, and autonomous driving. It is also suitable for applications that prioritize power efficiency and compact size, such as wearables. Furthermore, its open software platform enables developers to innovate and generate new applications with much less effort.
"Three significant enhancements distinguish our WiseEye2 from our first generation," Chen elaborates. "First, the power architecture. We implemented a multilayer power management mechanism and a multicore architecture to cater to various use cases. This results in a 32 times improvement in inference speed and 15 to 50 times better inference efficiency compared to our first generation."
In terms of use cases, WiseEye1 facilitates person detection, face detection, or object detection. WiseEye2 designed with enhanced AI capabilities, takes a step further by not only detecting the presence of a person but also discerning their actions or activities.
"By detecting facial landmarks, we can gauge the engagement level or even the emotions of the person in front of the camera. Additionally, we can conduct more precise object analysis while consuming less power," stated Chen. "WiseEye2 represents a significant breakthrough in low-power AI computing."
"Other critical characteristics of WiseEye2 are the embedded industry-leading security and privacy. " We incorporated PUF—Physical Unclonable Function—technology in WiseEye2" elaborates Chen. "This signifies that each chip possesses its own unique ID, which we utilize to create an encryption key safeguarding the software and ML features embedded on the chip. For privacy, WiseEye2 takes the image as the input to AI models but only transmits the metadata result as output. This prevents the system from acquiring the user’s image data, greatly improving user privacy."
Last but not least, WiseEye2 boasts a developer-friendly toolchain and robust ecosystem support. Designed with Arm CPU and NPU framework, Himax can utilize the toolchain to aid customers accommodate all available open-source AI models in the market.
"Allow me to illustrate some examples based on some real customer requests," says Chen. "For instance, in the case of a doorbell or door lock, the battery life needs to last at least six months. Our solution must support a 24-hour sentry mode, continuously monitoring the door for approaching individuals, day and night, seven days a week. Under these conditions, customers expect the solution to operate for six months without recharging."
"Another scenario involves the water meter. Leveraging our image sensor and AI processor, we can extract data from traditional water meters, whether it is digit-type or pointer-type, and wirelessly transmit only the required data, the metadata, to the cloud for follow-up analysis. In this case, most customers anticipate a battery life of eight years, assuming data readings once per hour," Chen explains. "In battery-powered applications, energy sustainability has become an essential requirement and that’s the design philosophy behind Himax WiseEye that we prioritize ultralow power consumption while maintaining robust computer vision capabilities."
Integrated solution
Numerous customers, including OEMs and ODMs, express the desire to incorporate AI functionalities into their products to enhance user experiences. However, most of them lack the necessary expertise to do so.
"Himax also offers turnkey solutions that include an ultralow power AI processor, always-on image sensor, and a comprehensive software stack equipped with pre-trained machine learning models," explains Chen. "Our WiseEye solution significantly reduces the high entry barrier associated with system development for deploying computer vision and machine learning AI capabilities to edge devices. This reduction encompasses the elimination of efforts in software, hardware, and camera design and development, as well as procurement and module manufacturing. Additionally, we will continue to roll out versatile AI vision features to further enhance the value of our products and expand the reach."
Award winning AI product
Himax’s WiseEye2 is one of the recipients of the 2023 Best AI Product award at the recent EE Awards Asia.
"While generative AI and AI related are hot topics right now, however, it also followed by the challenges posed by the high-power requirements of robust processors or cloud connections, resulting in rising costs, increased power consumption, deployment limitations, and privacy concerns in cloud computing. From this perspective, Himax’s proprietary WiseEye AI solution assumes a key role, particularly suited for resource-constrained, battery-powered endpoint AI applications, where it features local inferencing tinyML vision AI capabilities with ultralow power consumption, a compact form factor, and built-in industrial-grade security and cryptography engines. We are very happy to get this award and be recognized by the judges and engineer subscribers of EE Times in Asia," says Chen. "At the same time, we want to express our gratitude to EE Times for providing a platform to spotlight tinyML, especially emphasizing the significance in endpoint AI. to let designers and engineers know that we have industry-leading ultralow power AI solution to help them achieve a better user experience that reduce the entry barrier for development and enable a variety of applications."
Product showcase at CES 2024
Himax will be joining CES 2024 this month to highlight its latest product developments and AI innovations over the past year.
"One is the AR/VR edge tracking solution with our WiseEye2, with our in-house or third-party algorithm," says Chen. "We will also demo the door lock, which we already have in production with our customer. The uniqueness of this demo is that Himax’s vision AI and ultralow power consumption empowers door lock to realize 24/7 real-time monitoring and consecutive snapshot. All these are taking place while the SoC remains powered off, significantly reduces the overall power consumption as well as the cost."
The company will also showcase automotive applications, such as touch solutions, industry-first LTDI for pillar to pillar, large-sized automotive display at CES 2024. "This is to show how we can enrich the user experience in the smart cockpit," says Chen.
The year ahead
There remain numerous uncertainties across the world. However, from a technological standpoint, AI will undoubtedly remain the main theme in 2024 and beyond, evidenced by numerous vendors and leading cloud service providers intensifying their focus on developing solutions and services tailored to this sector.
"We’re observing a notable increase in customers seeking easy-to-adopt low-power and cost-effective AI solutions. Our dedication remains steadfast in advancing our AI capabilities and expanding our offerings to elevate human-machine interfaces, striving for an enhanced user experience, and to seamlessly integrate AI further into our day to day lives."