Chips&Media video IP enters Mobilint production

Chips&Media has confirmed that Mobilint has begun mass production of AI semiconductor devices incorporating its hardware video codec intellectual property, extending an established technology licensing relationship into processors intended for robotics and other embedded AI applications.

The South Korean semiconductor IP developer supplies video compression and decompression technologies that can be integrated into system-on-chip designs. Mobilint, which develops edge AI processors, has licensed several generations of its video codec IP as part of a wider collaboration combining multimedia processing with dedicated neural-network acceleration.

The production milestone was confirmed on 4 November 2026. Chips&Media identifies humanoid robotics and other physical AI equipment among the target markets, where semiconductor devices may need to process information from multiple cameras while supporting inference and control software.

The licensing relationship initially included H.264/AVC, H.265/HEVC, and AV1 technologies. It has subsequently expanded to cover newer codec standards and developments, including AV2 and H.266/VVC.

The wider licensing scope should be distinguished from the capabilities of an individual production chip. Chips&Media has not published a complete list of the codecs, profiles, resolutions, or concurrent processing configurations enabled in each Mobilint device entering production.

Hardware video processing is important in embedded systems that acquire information from several image sensors. A camera produces a continuous stream of pixel data, with the amount of information increasing according to resolution, frame rate, pixel format, and the number of connected cameras.

Multiple high-resolution streams can place substantial demands on an embedded processor’s memory system. Moving uncompressed image data between acquisition hardware, memory, processing engines, and communications interfaces consumes bandwidth and electrical power.

Video compression reduces the amount of data needed to represent image sequences by exploiting spatial and temporal relationships between pixels and successive frames. Established codecs combine prediction, mathematical transforms, quantisation, and entropy coding, although their detailed implementation varies between standards.

Software execution of these operations can consume considerable processor resources, particularly when several streams must be encoded or decoded simultaneously. A dedicated video processing unit performs supported operations in hardware, allowing general-purpose cores to concentrate on application software and other tasks.

Chips&Media’s semiconductor IP provides such hardware functions for integration into a larger chip. The customer is responsible for combining the video processing unit with the processor, memory system, communications interfaces, and other hardware blocks required by the completed design.

Mobilint’s edge AI architecture also uses neural processing resources intended to accelerate machine-learning inference. Combining video processing and inference within a single system-on-chip can provide a route for handling camera information close to its point of acquisition.

Robotic perception systems introduce requirements beyond conventional video recording. Camera data may need to be synchronised with other sensors, processed through computer vision algorithms, and analysed by neural networks before the controller can determine an appropriate response.

The processing architecture must therefore manage acquisition timing, memory transfers, computation, and the movement of results between hardware blocks. The number of cameras supported is only one element of the system’s performance.

Chips&Media identifies robotic systems incorporating between eight and twenty camera channels as examples of demanding potential applications. That range represents an application scenario described by the supplier, not a verified specification for a particular Mobilint production device.

The company also states that its video processing unit can achieve 8K resolution at 60 frames per second using a single core. This is a supplier-reported capability of the relevant IP rather than an independently measured result for every device incorporating the technology.

The effective performance of a production system will depend on the selected IP configuration, codec, frame format, memory bandwidth, clock frequency, and other implementation characteristics. Simultaneous handling of several video streams may require additional processing resources and careful management of shared memory access.

Video compression and neural inference also serve different functions. Compression can reduce the volume of information stored or transmitted, while inference extracts information such as object classifications or other analytical results from suitable image data.

In some applications, inference may operate on images before compression to avoid introducing distortion into the model’s input. Other systems may decode compressed information before analysis, depending on their communications and storage architecture.

These choices affect latency, memory demand, image quality, and power consumption. Integration of the video processing IP provides useful hardware capabilities but does not determine the complete perception pipeline.

Chips&Media also identifies functional safety as a factor in Mobilint’s technology selection. Its relevant IP has achieved ISO 26262 ASIL B certification, providing supporting development evidence for applications that require automotive functional safety processes.

That certification concerns the specified intellectual property and its development requirements. It does not automatically establish that a complete Mobilint processor or a robotic system incorporating the device meets ASIL B requirements.

System-level safety engineering must account for the intended application, diagnostic functions, fault detection, software behaviour, and the response to failures affecting the complete electronic assembly.

Chips&Media and Mobilint have previously worked together on multimedia processing technology. Mobilint has also developed its Regulus AI processor, which has been associated with a South Korean launch-vehicle programme.

The earlier aerospace application forms part of Mobilint’s technical background and should not be confused with the new mass-production milestone or treated as evidence that the current production devices have been qualified for every robotics environment.

Dong-joo Shin, chief executive of Mobilint, said the companies intended to combine Mobilint’s AI semiconductor technology with Chips&Media’s established video IP to support next-generation robotics applications, including humanoid systems.

The companies plan to extend the partnership into frame-buffer compression and data compression intellectual property. Frame-buffer compression can reduce the amount of image information transferred between processing units and memory, particularly where the same image buffers are accessed repeatedly during a processing sequence.

Reducing unnecessary memory traffic can lower bandwidth requirements, although any improvement in power consumption or overall performance depends on compression overhead, memory architecture, and the workloads involved.

Chips&Media has not disclosed the production volume, licensing income, final processor configurations, or customer shipment commitments associated with the Mobilint programme.

The announced milestone establishes that licensed video IP has progressed into Mobilint semiconductor production. Further processor specifications and customer deployments will determine how those capabilities are used in commercial robotics and multi-camera equipment.