Summary

• Ubitium has raised $14m in additional seed funding, bringing total seed investment since 2024 to $18m.
• Its UB108 prototype uses 256 processing elements and has demonstrated signal processing, AI inference and control on the same chip.
• The company plans a larger UB410 processor and customer evaluation units in 2027, initially targeting radar and audio systems.

German semiconductor startup Ubitium has secured $14 million to develop a commercial processor designed to handle sensing, AI inference and real-time control within a single reconfigurable architecture, following successful demonstrations using its first manufactured chip.

The Düsseldorf company is targeting autonomous equipment that must interpret sensor information and respond to changing conditions without relying on continuous connections to external computing infrastructure. Its proposed architecture combines functions often assigned to separate processors, potentially reducing the number of components and development environments required inside machines such as drones, robots and industrial equipment.

Announced on 13 October, the seed investment was led by Ibex Investors, with OpenOcean joining existing backers KBC Focus Fund, Runa Capital and Inflection. Florian Seibel, co-founder and co-chief executive of European defence technology company Quantum Systems, also participated personally. The financing brings Ubitium’s total seed investment to $18 million since it was founded in 2024.

Although the company is working towards a commercially available product, it has already manufactured and tested silicon demonstrating the central functions of its design. That distinction gives the funding a more developed technical basis than an architectural proposal, while leaving important questions about performance, production and customer adoption unresolved.

A common architecture for different workloads

Autonomous machines typically perform several types of computing at once. Sensors generate information requiring signal processing, AI models interpret that information, and control software determines how the equipment responds. These workloads place different demands on processing hardware, particularly when a machine must operate within strict limits on power consumption, weight and response time.

Equipment manufacturers can combine central processors, digital signal processors and specialised AI accelerators to meet those requirements. Some products integrate these functions within a system-on-chip, while others distribute them across several components, depending on the capabilities needed and the cost of the resulting design.

Moving information between processing units may introduce requirements for memory transfers, communications and coordination between different software components. The resulting complexity can increase engineering effort, particularly where manufacturers must maintain separate development tools and adapt applications to several hardware architectures.

Ubitium’s approach uses a coarse-grained reconfigurable array comprising programmable processing elements. Rather than relying exclusively on fixed-function hardware for each workload, the company aims to allocate computing resources across signal processing, AI inference and control functions as operating requirements change.

The architecture incorporates unified memory and tools based on the open RISC-V instruction set. Ubitium argues that this combination could simplify software development by providing a common computing foundation for tasks that would otherwise require several specialised processors.

Its proposition remains dependent on the efficiency of the completed design. Reconfigurable hardware can provide flexibility, but must still execute different workloads within acceptable power and timing limits. The performance of a system running several functions concurrently may differ from benchmarks that examine those functions individually.

Standard development tools can reduce some integration work, although RISC-V compatibility does not mean applications written for any other processor will run without adaptation. Compiler support, development libraries, debugging capabilities and customer-specific optimisation all influence whether a new architecture can be adopted economically.

Working silicon establishes the first engineering milestone

Ubitium’s initial processor, the UB108, contains 256 processing elements and was manufactured using Samsung’s 8nm semiconductor process. The company says the device has demonstrated signal processing, AI inference and real-time control operating on the same reconfigurable array.

That evidence establishes that the functions can execute on manufactured hardware, moving the development beyond simulation. However, the funding announcement does not provide independent benchmarks comparing the processor’s energy consumption, processing latency or cost with established alternatives under equivalent operating conditions.

The company also reports 17 granted patents covering its core technology, alongside further applications. Such intellectual property may support the protection of its design, but patents do not establish commercial performance or the ability to manufacture processors competitively at high volumes.

Following the UB108 demonstration, Ubitium plans to tape out a larger commercial processor designated UB410. The proposed design contains 4,096 processing elements, substantially increasing the resources available within the reconfigurable array.

Ubitium intends to deliver evaluation units during 2027, allowing prospective customers to test the technology against their own requirements. The timetable remains a development plan rather than a confirmed start of volume production.

Tape-out involves submitting the completed chip design for manufacturing, after which samples must undergo testing and validation. The process can reveal issues involving power consumption, manufacturing yield, timing or compatibility that require further engineering work before a device is ready for commercial deployment.

Manufacturers considering a new processor must also evaluate how reliably its supplier can deliver components over the life of their products. Equipment used in industrial and other demanding applications may remain in service for years, making product availability, software maintenance and the qualification of hardware changes important procurement considerations.

Radar and audio provide initial commercial tests

Ubitium says its initial customers are developing radar and audio systems, applications in which signal processing and AI increasingly need to operate together. These markets provide opportunities to assess the architecture against established computing requirements before extending it to more complicated autonomous machines.

A radar system, for example, processes information from received signals before software can identify objects or interpret movement. Combining those activities on one processing architecture could reduce some data movement and integration work, although the benefits would depend on the sensor design, algorithms and operating conditions.

Audio applications create different processing demands involving continuous incoming signals, recognition and other calculations. A common architecture could offer advantages to manufacturers developing several product types, provided the same hardware can meet their distinct performance and energy requirements.

The longer-term opportunity involves autonomous equipment carrying out sensing, decision-making and physical control without depending on remote servers. Such systems may need to operate when communications are unavailable, particularly in industrial environments or applications where connectivity cannot be guaranteed.

Safety and reliability requirements become more demanding when computing outputs influence physical movement. Manufacturers must consider how quickly a device responds, how errors are detected and whether the system continues to operate predictably when individual components or software functions fail.

Ubitium’s architecture is therefore being developed for a market in which processor flexibility needs to be assessed alongside predictable behaviour. Its current demonstration does not establish that the planned UB410 satisfies the qualification requirements of any particular safety-critical application.

European investment in physical AI has also extended into the companies building the machines themselves. Techopia recently examined RobCo’s expansion in industrial robotics, where the commercial challenge concerns introducing autonomous capabilities into manufacturing operations. Processor development addresses a different part of that process, but the wider market depends on hardware, software and operational systems working reliably together.

With $14 million in new funding and working first silicon, Ubitium can continue developing a commercial processor around its reconfigurable design. The UB410 evaluation programme will provide the next evidence of whether combining several computing functions delivers the performance and engineering advantages required by prospective customers.