アナログ・テック Product Lineup

The AT-IPCFG series from Analog Tech deploys image AI and edge AI for harsh environments where computers struggle, such as safety management at construction, agriculture, and surveying sites, inspection sites for roads, tunnels, and bridges, vehicles like unmanned transport vehicles, and environments with high temperatures, low temperatures, high humidity, and dust. Equipped with a ruggedized housing that enhances environmental performance, it includes AI chips and AI modules for inference. It provides low-latency, high-speed image AI processing on edge devices, including high-resolution image processing, video processing from multiple cameras, real-time image processing, and AI inference processing. Custom solutions and OEM supply tailored to customer applications are available, so please feel free to consult with us.

The AT-IPCWG series is a compact wall-mounted industrial PC that can accommodate a variety of NVIDIA GPUs ranging from mid to high-end. Customization and OEM options are available to suit customer needs, so please consult with our consultants. 【Features】 ■ Equipped with NVIDIA Geforce/T1000/T4/Quadro Thanks to excellent cooling performance, it can support various NVIDIA GPUs. It is compatible with the relatively affordable Geforce series for gaming, the T4 for data centers, and the Quadro series for image processing and simulation. You can also choose fanless GPUs like the T4 or passive thermal types of Quadro. ■ Expandable with various functional boards including image input boards A chassis designed for enhanced expandability is available. It supports various input and output options tailored to your needs, such as expanding inputs for Camera Link and CoaXpress. ■ Installable on-site with dust filter Standard equipped with a dust filter. It boasts excellent environmental resistance and can be deployed in locations where rack mounting is difficult. 【Further features continue in the section "Basic Information" below】

Our company manufactures industrial PCs and servers tailored to customer requests, and we have prepared base models that embody various customer demands and the latest technologies in advance. If the specifications match, you can use them as they are. If the specifications do not match, we can customize the parts of the base model that do not fit to shorten the specification design process and reduce the lead time for delivery to customers. We will start by confirming base chassis that are close to the customer's required specifications. If you have a chassis that resembles your required specifications, please feel free to consult with us. For the main lineup of base models, please refer to the basic information below.

■What is an AI Chip■ An AI chip is a semiconductor designed to accelerate AI computational processing. In image processing and AI tasks, GPUs are becoming almost standardized for training purposes. However, the specifications differ between training and inference, and inference can be achieved using not only GPUs but also FPGAs and ASICs. In particular, inference for edge applications is implemented in devices that process AI on-site alongside IoT, necessitating considerations for adoption based on not only "computational performance" but also "low power consumption," "wide operating temperature," and "long supply duration" for embedded implementations in devices. ■Introduction of AI Chips Suitable for Industrial Applications■ Hailo's AI chip boasting low power consumption and high inference processing capability: 'Hailo-8 AI Processor' The Hailo-8 processor is a chip specifically developed by Hailo for AI processing. Among AI chips, it has the following features that make it particularly suitable for industrial applications: (Feature 1) AI processing with high power efficiency! Power efficiency of 3 TOPS/W (Feature 2) High environmental durability! Operating temperature range of -40°C to 85°C (Feature 3) Planned for long-term supply! Expected supply duration of up to 10 years for the chip alone.

■Challenges of Implementing Image AI and Edge AI■ Recently, the development of various image AI and edge AI applications has progressed, moving into the device implementation phase. In this phase, unlike in offices or data centers, devices are required to have stringent environmental durability, which presents a challenge for the implementation of edge AI. On the other hand, the data processed at the edge continues to increase in volume and real-time demands, leading to a linear rise in component performance, resulting in extreme heat generation from CPUs, GPUs, and AI chips, which adds to the challenges of device implementation. In edge AI, to enhance processing performance, it is crucial to address heat dissipation from high-heat-generating components such as GPUs and AI chips while ensuring extremely strict environmental durability. This involves temperature control, vibration mitigation, and mechanical design, which are important factors for implementing image AI and edge AI applications in various settings. ■High Environmental Durability Edge AI Platform■ Analog Devices provides a platform for implementing AI in harsh field environments such as 'high temperature/low temperature, water, high humidity, and dust' by equipping industrial PCs with AI chips that excel in environmental durability.

We have packaged a development environment to easily start developing AI applications using the Hailo-8 AI Processor. The evaluation development kit includes PC hardware equipped with a Hailo-8 Mini-PCIe module (with a performance value of 13 TOPS) as the execution environment for running Hailo-8. ■ Deployment toolchain compatible with industry-standard frameworks It supports model conversion from industry-standard frameworks to Hailo format, converting the model's bit configuration to Hailo's configuration using the latest quantization algorithms. It allocates hardware resources and user network resources to the physical resources of the Hailo device. The model is compiled into a Hailo binary, the binary is loaded, and inference is executed on the Hailo target device. The SDK supports both standalone inference, which allows direct access to the device, and TensorFlow integrated inference, which facilitates integration with existing environments.

■Equipped with two 'Hailo-8 AI processors' in a single fanless chassis■ The 'AT-IPCFG004' is an industrial computer with excellent environmental performance, featuring two high-performance AI chips designed for industrial applications. It can simultaneously apply different learning models, such as area recognition and object recognition, as well as object recognition and pose estimation, to input sources, thereby expanding AI inference capabilities. ■Equipped with two high-performance AI chips from Hailo, supporting multi-AI algorithms■ In AI deep learning, depending on the 'types of data sources' being handled, such as images, audio, and time series, as well as the 'analysis purposes' for what you want to learn from that data source, suitable algorithms are selected and applied accordingly for tasks like classification, depth estimation, object recognition, face detection, and pose estimation. Generally, in edge inference, one learning model is executed for one image source. With the AT-IPCFG004, by incorporating multiple Hailo-8 processors, it has become possible to 'simultaneously apply multiple algorithms to the source.'