Designing the next generation of AI accelerators — from low-power edge chips to brain-inspired neuromorphic architectures — where hardware and intelligence are co-engineered from the ground up.
About the Division
The AI Hardware Research division at Presear Softwares investigates the architectural and circuit-level foundations needed to make AI computation faster, smaller, and more energy-efficient. As Moore's Law approaches physical limits, the next leap in AI performance will come from purpose-built hardware — and that is precisely where our research is focused.
We work across the full hardware stack: from algorithm-hardware co-design and compiler-level optimisation to custom accelerator architecture and embedded system integration. Our teams collaborate with chip designers, cloud providers, and edge device manufacturers to translate research prototypes into manufacturable silicon that meets real deployment requirements.
Central to our mission is sustainability — designing AI hardware that delivers high throughput while dramatically reducing energy consumption, thermal output, and material cost. From neuromorphic chips inspired by the human brain to scalable tensor processing units, we are shaping what AI compute looks like beyond the GPU era.
Research Focus
Six interconnected areas where Presear's AI Hardware Research division is defining the future of intelligent computing silicon.
Designing domain-specific accelerators — from tensor processing units to sparse matrix engines — that maximise AI throughput per watt across training and inference workloads at cloud and edge scale.
Building compact, ultra-low-power inference hardware for deployment in IoT sensors, wearables, industrial controllers, and autonomous systems where connectivity and battery life are primary constraints.
Researching spiking neural network hardware, memristive crossbar arrays, and event-driven computation paradigms that mimic biological neural circuits for efficient, always-on intelligence at minimal power.
Jointly optimising AI model architectures with underlying hardware topologies — using neural architecture search, quantisation, pruning, and compiler-aware model design to close the gap between algorithm and silicon.
Developing power management strategies, approximate computing techniques, and thermal-aware scheduling that cut the energy cost of AI training and inference without sacrificing model accuracy or reliability.
We work with semiconductor companies, cloud providers, and system integrators to co-design and validate AI hardware solutions from research prototype to production silicon.
Presear AI Hardware Research partners with semiconductor companies, hyperscalers, and product teams to co-design silicon that makes intelligent systems feasible at every scale.