Singapore-headquartered semiconductor startup Symatics unveiled its Morphic Processing Unit (MPU) architecture at the Business & Philanthropy Forum 2026 in Singapore, introducing a new approach to how AI computations are organised and executed.

The company said the MPU reorganises AI computing around row-column parallelism, built for interactive AI, complex reasoning, and AI agents that depend on how fast a single request can be completed. Symatics will develop the architecture in collaboration with research and industry partners in Singapore, supporting the company's global ambitions while contributing to the region's AI computing capabilities.

Rethinking how AI processes information

According to Symatics, one challenge in AI computing is the repeated movement of information between memory and processing components, which can slow execution. The company says its architecture is designed to address this bottleneck.

The MPU maps the parallel unfolding of a computation, and the dependencies between its steps, onto a row-column array, so data supply is matched precisely to where each calculation takes place. Intermediate results stay on the chip and are used immediately nearby, while computing state is retained and continuously updated. This reduces unnecessary external memory reads and writes and data redistribution, so each stage of an algorithm triggers the next and keeps running.

A compiler plans where computations take place, how information moves between them and the order in which operations are executed. By determining these steps in advance, the architecture aims to reduce the time and resources spent coordinating tasks during execution.

Scaling up individual AI requests

Symatics said the architecture also aims to improve the speed of individual AI requests, rather than focusing solely on the total number of requests a system can process.

When a single stream is limited by bandwidth or compute resources, the MPU can spread the same request across more MPUs, expanding computing and data supply together to raise single-stream speed. It is designed to give AI services a competitive edge where response time matters most. A single scale-up domain allows operators to raise per-request token rates, not just aggregate throughput, as they add hardware.

Strengthening Singapore's role in AI computing

Under its open architecture, Symatics will support partners in customising chips and systems, and share reference designs and implementation experience to build an open ecosystem, followed by a configurable B2B infrastructure platform.

With research and industry partners, Symatics will build a Singapore implementation standard and demonstration projects that validate the path from model compilation to deployment. It will connect Asia-Pacific manufacturing, strengthen local design services and offer customised chip and system engineering for enterprises and national-level customers, building original intellectual property and engineering talent in Singapore.

Company background

Symatics is a Singapore-headquartered semiconductor startup developing the Morphic Processing Unit, a computing architecture designed to improve the performance and efficiency of AI workloads. The company said it has engineering teams in Singapore, Taiwan and Toronto and is advancing its architecture through research and engineering development, with ambitions to serve the global AI infrastructure market.