The Emerging Applications of AI in Test and Measurement Automation in Today’s Chip-eat-Chip World 

As the digital facets of our lives continue to grow, so does the importance of the ubiquitous chip. Semiconductors, now poised to be the hero of a trillion-dollar industry by 2030, carry the heavy responsibility of delivering the fast-paced, complex technologies that the world has become accustomed to. With that, semiconductor testing, a vital part to ensure it meets stringent standards of quality and function, has also grown to be complex, and fast-paced. This is where the exploration of AI in Test and Measurement Automation becomes extremely important in trying to address this growing urgency in the market. 

Unwrapping the Existing Gaps in Test and Measurement Automation 

In Test and Measurement there is a typical workflow followed by most semiconductor companies: 

1. Planning – Define your requirements and create specifications 
2. Set Up – Identify hardware and software for your projects 
3. Development – Write test scripts and sequences and complete debugging 
4. Data Collection – Where you execute your programs 
5. Analysis and Reporting – Generate correlations and reports 

The industry uses automation solutions by vendors or in-house, to help manage these different phases. But there are two problems that you find with this approach. 

1. Broken Automation – These workflows are not connected. The software and tools of one workflow does not actually work with the software and tools of the next workflow. Here, engineers act as connectors. The only other alternative is to take up a large standardization effort to bring all these workflows together. 
2. Reactive Tools – The software or tools available for workflows are reactive and only help manage the workflows and do not actually provide support to the engineer in their core execution activities. 

At Soliton, we are exploring the use of Applied AI in bringing proactive intelligence to support the work of these engineers and improve their engineering experience. AI can be added to existing tools to achieve this. 

So, what is relevant, where? 

Two years ago, AI was primarily used in Test and Measurement to tackle high volumes of data for anomaly detection and yield predictions. Traditional AI also has been widely used in parameter optimization and to perform waterfall analyses.  

An AI CoPilot to Help Write Test Programs 

With generative AI, there’s a lot of specifications and code generation work in verification and validation that can fit well into generative AI’s LLM models. This also includes writing code for manufacturing programs – all tasks ordinarily engineer-intensive and repetitive. 

So, if you were an engineer tasked with creating test programs you would have  

1. a list of requirements or criterion that your test must satisfy 
2. a bench set-up that consists of your automated framework, instrument and DUT drivers, libraries that come with it and some pin information. 

The engineer’s next two steps would be to  

• draw a test methodology with relevant formulas, and following procedures prescribed by the company. 
• and create a Test Program. To write code befitting the framework and pins with measurement plugins for relevant drivers and hardware to form a test sequence. 

These are the two steps that an engineer could use an AI CoPilot to support her/ his work.  

The Gen AI Co-pilot can generate a test methodology based on the requirements and with AI in the loop the engineer can review and edit or accept the AI results. Once the test methodology is ready, the engineer can again use an AI CoPilot and move on to then generate code from the finalized test methodology, in the same way – review and edit with AI in the loop and then accept a final test program. 

An AI Chatbot to Access Several Data Sources 

Now what if you wanted to go even further back into the workflow? If you were a validation engineer, you need to access different types of data such as specification, test methodologies, measurement data and so on. All of this data exists in different formats and sources. It needs to be carefully studied to ensure compliance and to ensure the product reaches a high standard of quality.  

Since this is data that lies within the organization itself, AI can be used as a layer on top of the organization’s internal data to help the engineer search and summarize relevant data prompted by a query. And all this with the data being exactly how it exists within the organization – in its different sources and formats. 

So, with a simple chat interface, you can enter a query related to insights or data and the Chatbot can search for relevant information and summarize an answer to your question. The data sources the chatbot has access to are visible, retrieving both structured and unstructured formats.  

Analytics and Reporting with AI 

With access to various sources of data comes analytics and insights generation. AI has been prevalent in this space for at least a decade now – predictive maintenance, automated root cause analysis, anomaly detection and yield prediction. 

But perhaps it has not been as accessible as it is today. The same AI Chatbot can now provide sophisticated analytics, given it has in its database all the relevant algorithms as well. So say you recall a set of data through a query and then ask the chatbot to search for the anomaly. AI would pick an algorithm that would match the query and apply it to the data and run the algorithm in the back end and gives you the data analytics that you wanted.  

Similarly, you could pick a certain data visualization option that suits the data set that you want to analyze and then extract these for your use. 

Scratching the Surface 

We honestly do believe that all of this is just scratching the surface and there’s a lot more to explore and uncover. But one thing you can be sure of – AI is going to take center stage in the demanding field of Test and Measurement Automation. It is a breakthrough that we believe can redefine the work that test, validation and verification engineers take up in the future.