Failure Analysis for AI and Advanced Semiconductors: Locating the Most Difficult to Find Defects Using Ultra-high Frequency Nanoprobing

Electrical Failure Analysis for AI and Advanced Semiconductors: Locating the Most Difficult to Find Defects Using Ultra-High Frequency Nanoprobing

 

Rarely does a news cycle pass without a reference to artificial intelligence (AI) and how it is impacting yet another industry. Whether it is automating business operations, transforming manufacturing, or enabling new creative content it is seemingly everywhere. However, one hears less often about the hardware that has enabled AI.

The development of AI has necessitated the creation of new specialized hardware that integrate with the computer’s central processing unit. These include graphics processing units for parallel computing, tensor processing units for high-speed machine learning, neural processing units for on-device tasks like computer vision and natural language processing, and floating-point units for ultra-fast arithmetic calculations.

This new AI hardware also has intensive energy demands. This power requirement has been the primary driver behind two additional advanced semiconductor technologies: gate-all-around transistors and back-side power delivery networks. These new devices and architectures add complexity, and when electrical failures occur, they can be exceedingly difficult to diagnose.

Transient failures are among the most difficult electrical issues to detect in these chips. They result from faults that produce time-dependent errors, which standard current and voltage measurement techniques often miss—especially when the failures depend on sequential switching. Diagnosing these issues therefore requires measurement methods that capture signals over time. Moreover, these methods must be fast enough to accurately track signals at typical operating frequencies of up to 6 GHz.

However, existing time-dependent techniques at the nanoscale are significantly slower, with response frequencies limited to about 0.1 GHz. Tiptek is addressing this gap by developing novel probe technology capable of nanometer-scale measurements at several gigahertz. This advancement will make it possible to locate transient failures in the most advanced semiconductors, including those used in AI systems.