SpikePro

Rapid advances in artificial intelligence have led to powerful models and algorithms which drive conventional computing hardware to its limits regarding energy efficiency and speed. The EIC SPIKEPro project develops a novel neuromorphic chip technology, enabling spiking neural networks (SNN) to process data in more efficient ways. SPIKEPro proposes a science-towards-technology breakthrough by combining low-energy electrical and photonic neurons with memristive weighting elements into a joint spiking neural network on chip.
This hop-on project significantly accelerates and facilitates SPIKEPro by providing solutions for accurate control, interconnection and tuning of the analog SNN, not planned in the original project. The hop-on partner INL will develop specifically tailored Low-Power Analog Mixed-Signal CMOS control circuits and connect those to the analog SNN through a high-density chiplet interconnection process at wafer scale (200-mm). This will enable automated control, training and test of the new SNN technology and at the same time validate neuromorphic technology up-scaling and elevate the TRL. Furthermore, novel integrated circuits will be designed to tune the performance of electrical spiking nodes, which is exploited as additional means for efficient training.
As a consequence, the SPIKEPro demonstrators can use the hop-on control circuits and benefit from fully integrated automatic control algorithms and hardware-level feedback loops. This will maximize the impact of SPIKEPro’s scientific activities and provide a concrete route towards prototype level technology. This project will be a timely opportunity in creating a new semiconductor chiplet packaging platform for neuromorphic computing in a PT-North European partnership, leveraging recent investments of PT in the Chips Act. It will reinforce the country and the region’s national semiconductor ecosystem identified as of strategic economic importance, specifically in the advanced packaging area.