Tria introduces compact module for industrial edge AI

Tria Technologies has introduced the C6C-PLH, a COM Express Compact Type 6 computer-on-module based on Intel Core Ultra Series 3 processors, combining general-purpose computing, graphics, and neural-network acceleration within a 95 × 95 mm embedded platform.

The new module targets industrial automation, robotics, machine vision, medical imaging, and advanced operator interfaces. Tria has developed the platform for equipment requiring substantial local processing capacity alongside industrial operating characteristics and long-term product support.

Processor options include configurations offering up to 16 CPU cores, integrated Intel graphics with up to 12 Xe cores, and a neural processing unit (NPU). The module also supports up to 96 GB of LPDDR5x memory and a range of high-speed interfaces, allowing the computing resources to be incorporated into application-specific embedded systems.

Edge computing equipment increasingly combines conventional application software with demanding workloads such as image processing, object detection, anomaly detection, and real-time monitoring. These functions can require substantial processing capacity while operating within the power, space, and thermal restrictions of an industrial enclosure.

Different computational workloads have different characteristics. A general-purpose CPU provides flexibility for operating systems, communications, and application logic, while a graphics processor can execute parallel mathematical operations and accelerate image-processing tasks.

A neural processor is designed for selected machine-learning calculations, enabling compatible models to execute without relying exclusively on conventional CPU instructions. Its practical contribution depends on the numerical formats, supported operations, memory requirements, and software implementation of the selected model.

The C6C-PLH combines these processing resources within one module, allowing developers to allocate work across the available hardware. A machine vision application might use the graphics subsystem for image preparation, the NPU for object classification, and the CPU for supervisory software and communication with surrounding equipment.

Such arrangements can reduce dependence on external computing resources when the available hardware provides sufficient performance for the intended workload. They do not eliminate the need to evaluate processing latency, memory bandwidth, and the interaction between concurrent applications.

Local inference can be useful in systems that require rapid responses or cannot depend on continuous cloud connectivity. However, deterministic control still depends on the software architecture, operating system, peripheral behaviour, and the allocation of shared processing resources.

Memory capacity is another consideration in embedded AI. Neural-network parameters, image buffers, intermediate calculations, and application data may all need to remain accessible while different processing engines are active.

Tria supports up to 96 GB of LPDDR5x memory, with in-band error-correction options. The available memory configuration depends on the selected module variant, while the effective performance of the complete system will be influenced by processor characteristics and competing memory demands.

In-band error correction provides a mechanism for detecting and correcting supported memory errors using capabilities associated with the memory subsystem. Its protection characteristics must be considered within the reliability requirements of the complete application, particularly where the equipment performs critical control or monitoring functions.

The module supports four independent display streams, extending its suitability to equipment incorporating multiple screens or complex graphical interfaces. Industrial operator stations, diagnostic systems, and medical imaging equipment may require simultaneous presentation of several information sources.

Integrated graphics can support display rendering and media processing without requiring a separate discrete graphics card. The achievable performance will nevertheless depend on the selected processor, display configuration, software, and thermal operating conditions.

High-speed interfaces include PCI Express Gen 4, USB connectivity incorporating USB4, and a 2.5 Gb Ethernet interface using Intel’s i226 controller. These provide connections for storage, external peripherals, network communications, and application-specific expansion hardware.

PCI Express can connect additional processing resources, acquisition hardware, or high-performance storage devices through the carrier board. USB4 provides another high-bandwidth interface option, although the available ports and their electrical implementation depend on the completed carrier-board design.

Tria also identifies support for Intel Time Coordinated Computing and Time Sensitive Networking technologies. These capabilities are intended to assist applications requiring coordinated processing and network communications, particularly in industrial systems where timing behaviour must be managed across several devices.

Time-sensitive networking introduces mechanisms for controlling the transmission of selected data across compatible Ethernet networks. Its effectiveness depends on the configuration of the network, the capabilities of connected equipment, and the timing requirements of the application.

Support for these technologies does not automatically guarantee deterministic behaviour from every operating system or software workload. The complete system must be configured and validated against its required communication and processing deadlines.

The C6C-PLH uses the COM Express Type 6 pinout, providing a standards-based interface between the processor module and an application-specific carrier board. The carrier board supplies the physical connections, power distribution, and peripheral interfaces required by the finished equipment.

Separating the processing module from the carrier board can reduce the amount of processor-specific hardware design required for an embedded product. It also creates opportunities to reuse a carrier-board architecture when moving between compatible computing modules.

Mechanical and electrical compatibility must still be established for each combination. Processor power consumption, cooling arrangements, firmware requirements, and the availability of individual interfaces may differ between module generations, even where the same nominal standard is used.

The C6C-PLH is available with industrial temperature options covering -40°C to +85°C and configurations intended for continuous operation. These environmental specifications apply to supported variants rather than every possible processor and cooling arrangement.

Thermal management becomes important when CPU, GPU, and neural-processing workloads operate simultaneously. The sustained performance available from an embedded system can be limited by the ability of its cooling arrangement to remove heat from the processor and surrounding components.

Tria offers evaluation carrier boards and cooling solutions to assist development and integration. These allow engineers to examine electrical connections, mechanical fit, software compatibility, and thermal behaviour before committing to a final carrier-board design.

Security features include Trusted Platform Module 2.0 support, providing hardware capabilities that can be used for functions such as protecting cryptographic material and establishing system integrity. The security of a completed product will also depend on firmware configuration, operating system maintenance, and the implementation of the wider software environment.

Operating system support includes Windows 11 IoT Enterprise LTSC 2024 and a Linux board support package based on Yocto. These options provide routes for equipment manufacturers to develop systems using different software environments and maintenance strategies.

Daniel Denzler, Vice President of Tria Technologies, said: “Industrial customers are not simply looking for more processing power; they need a platform that can sustain demanding workloads throughout long product lifecycles.”

Tria, an Avnet company, is positioning the module for embedded systems where future software maintenance and hardware availability are important alongside initial computational performance.

The company’s announcement does not establish a single performance benchmark for every processor configuration, nor does it disclose a universal price or lead time for the module family. Individual specifications and availability will need to be established for the selected product configuration.

Engineering evaluation can proceed through the company’s carrier-board and cooling options, with system integration focused on the processing resources, memory requirements, network timing, and thermal limits of the intended application.