Neuromorphic computing systems offer disruptive efficiency gains by a factor of 10-100 over conventional technology. These are achieved through a paradigm shift in computing hardware. The core innovation of neuromorphic chips are memristors, novel devices that both store and process data. As such, they offer the most efficient design for AI applications. Suppliers of initial neuromorphic chips include Intel and IHP. However, the available technology is currently inaccessible to potential users, primarily manufacturers of AI-enabled devices: they lack the expertise for use cases that is difficult to build up internally; they do not know how to integrate neuromorphic chips into functioning overall systems; and their applications cannot be migrated to neuromorphic hardware using known methods. This is due to the more complicated system design on both the hardware and software sides.
In this project, we facilitate AI-enabled device manufacturers to develop disruptive use cases on neuromorphic computing systems via a software development kit (SDK) and a hardware development kit (HDK). The SDK enables easy programming of AI applications on neuromorphic systems through compiler functions and simulates applications before neuromorphic chips are ready for production. The HDK integrates purchased neuromorphic chips into a working system, allowing users to validate neuromorphic hardware and serving as a development platform.
We have successfully demonstrated the disruptive potential of neuromorphic computing systems in the lab. For proof-of-concept, we developed an HDK with an integrated neuromorphic chip from IHP and tested an associated SDK. Now we evaluate the SDK and HDK with a prototype. To this end, we are developing a camera-based AI device with a pilot partner. With the prototype, we demonstrate that our SDK can solve a real-world application and that our HDK can be used to develop a market-ready product. The Innovation Sprint allows ICE to make a real technical and commercial leap forward.
