The aim of KI-EDA was to harness the tools of artificial intelligence (AI) for the rapid and reliable design of customised encoder and sensor chips. Such microelectronic systems are needed for novel intelligent, autonomous production systems that can securely analyse data in a decentralised manner to predict and improve manufacturing scenarios.

Conventional design methods are usually not sufficient for complex and individual designs

Encoder and sensor chips are used, for example, to detect rotary movements on machines and convert them into electronic signals for digital processing. These chips are customised for production machinery and processes and are becoming increasingly powerful and complex. The design and verification of the chips is reaching its limits with conventional electronic design automation (EDA) methods and is largely shaped by the experiential knowledge of design engineers.

New methods should reduce the development time of customised chips

This is why the KI-EDA project, which is being led by iC-Haus GmbH, has  worked on AI-supported methods of design automation and on a modular system for chip design and test. Individual functions should be able to be selected, combined and simulated quickly and cost-effectively in the form of function blocks with short design times and a low error rate, and transferred into an encoder or sensor chip customised for the planned application. On this basis it will be possible to implement new functionalities such as AI-based predictive maintenance solutions flexibly and quickly. Errors in the design and thus time-consuming redesigns are to be reduced in this way.

To this end, the partners iC-Haus and CENTITECH have researched and implemented intelligent functions and energy-autonomous solutions for the new microelectronic systems and characterised them in application-oriented demonstrators. This was intended to ensure the properties required for widespread and comprehensive use in I4.0 production environments. In addition, the partners ensure that the systems can be manufactured in high quantities.

IMMS researched on ML algorithms and smart models for a modular system for fast and safe chip design

IMMS researched machine learning (ML) algorithms for their applicability in chip design (EDA) as a basis for transfer into a novel designer assistance system to make design processes more efficient and faster. The new ML methods researched will primarily serve a fast design validation and thus increase design security. They are aimed in particular at model-based debugging and should reduce the complexity of working with IP libraries, i.e. with collections of existing circuit components or electronic blocks. New chips can thus be assembled in a modular system. In addition, IMMS researched and implemented „smart models“. For the first time, new algorithms enabled models to computationally check their own range of validity and thus exclude false-positive verification results.