AI for Engineering: Transforming Workflow

 

Using AI for engineering will improve performance before and after material breakthroughs.

2022 was the year of the AI renaissance. The rise of generative language chatbots has vastly reshaped what is possible with AI, and its adoption has become widespread. AI has been knocking on the door for some time, however: older readers may recall the headlines when Deep Blue, a chess-playing supercomputer, was able to beat world champion Garry Kasparov in 1997. While there have always been several attention-grabbing headlines for AI, its development has been more continuous and subdued than these highlights would let on. As AI continues to grow, so too does its suitability: using AI for engineering purposes has become commonplace and will continue to grow in prominence with hardware and algorithm improvements.

The Impact of AI by Engineering Discipline

How Using AI for Engineering Benefits Electronic DFM

To be clear, AI is nothing new. For years, the development of products and processes has heavily utilized different heuristic models of organizing large amounts of data for a response that can operate without additional operator input. Take neural networks used to train computers for defects or deviations in manufacturing: while there are many different categories of neural networks, they come down to pattern recognition. For example, if you show a training model an image of a bird enough times, it can begin to parse out some characteristic features. Depending on the level of training, showing this model a bat may cause an incorrect categorization, as it superficially resembles a bird in many aspects (while simultaneously differing tremendously).

Not all training models are on visual data. Other methods of applying weights or probabilities to data run through the model assign some correctness/incorrectness or preferred/disfavored evaluation based on its training. AI is less a single method for harnessing data and more an open-ended description of different systems for using data to exhibit better design elasticity. Some of AI’s most relevant capabilities to electronic design include:

• Circuit analysis – AI can incorporate multi-dimensional evaluation to improve a board layout. Consider power topologies: most basic topologies are well-described regarding placement between components, values, design of copper features, and more. AI can iterate through a design with a system that rewards or incentivizes based on improving performance (e.g., reducing ripple current, oscillation, etc.)
• Chip optimization – Power and size are the two most substantial considerations of high-performance ICs. For years, manufacturing and material improvements were able to shrink processes. Still, IC design is now butting up against some physical realities that have slowed the growth prediction of Moore’s law (the observation of chip transistor counts doubling every two years due to improvements in manufacturing). Chip designers can use AI to make subtle improvements to the die layout to extract further shrinkage and power gain.
• Manufacturing support – Whether at the chip or board level, more complex designs have made inspections more difficult yet necessary. Defect analysis needs to use multiple data acquisition modes and distinct models to exhibit human-accurate (or better) decision-making. One area where AI can have an outsized improvement on human performance is during automated optical inspection (AOI), which takes a known good board as comparison against all other boards to detect manufacturing errors. A convolutional neural network (CNN) partitions the board into small segments and applies filters that allow for discernment.

Subsets of AI and Their Impact on Engineering

The transformative impact of AI on design is undeniable. Advanced algorithms can find use in jumpstarting a project, optimizing decision trees, or enhancing design rule checks. The possibilities for AI are nearly endless due to the vastness of applications, with new paradigms continually opening up as technology and algorithms advance:

• Machine learning – ML analyzes data to identify trends and make reasonable predictions. It is heavily tied to probability and lends itself to system optimization by minimizing a loss function (crudely, the error between the prediction and the actual result). The learning mode can be supervised (with a known answer), unsupervised (without a known solution – generally performed after supervised learning), and reinforced (where a reward system incentivizes the algorithm’s performance).
  • Deep learning – A subset of machine learning that uses neural networks of various layers of depth to extract more complex features from data progressively.

• Data mining – Unlike machine learning, data mining analyzes data to uncover unknown knowledge. The Internet of Things (IoT) model attempts to improve process controls by harvesting data previously unused through sensors before transmitting it to the cloud for computationally intensive modeling.
• Machine perception – Computers can form rudimentary filtering by “sense” that mimics biological neural networks. 
  • Computer vision – Arguably the foremost application of AI before generative applications. Sensors use visible or infrared light to detect, track, recognize, and estimate visual stimuli – such as in autonomous vehicles.
  • Speech recognition – Voice-activated systems that usually rely on some amount of training by the speaker. 

• Natural language processing -Language is distinct in form and how individual users wield it. This form of AI must understand the complexity and nuance of large language sections and develop a coherent response. While there are issues with hallucinations – information the algorithm believes correct but is incorrect – chat models powered by NLP have rapidly gained prominence within the past months.
• Planning – AI can navigate branching decision trees to determine the best action. Unlike control systems with strictly defined outputs to known stimuli, planning involves operating in decision spaces where information gathering is necessary.
• Robotics – The system designed for autonomous or controlled machines. Often used in environments unsafe or uninhabitable for humans and tasks that require high precision and repeatability.

Ultra Librarian Supports AI with a Comprehensive Component Library

Using AI for engineering will continue to be paramount to driving innovation in the coming decades. Barring breakthroughs in materials that would support technologies currently incapable of realization, AI has significant potential to leverage existing systems further. Advancements in AI will rely on increasingly complex electronic systems where the room for error by design teams is narrower than ever. 

With Ultra Librarian’s catalog of millions of land patterns, symbols, and simulations alongside support for popular ECAD applications, design teams can place and route with total confidence and greater efficiency.

Working with Ultra Librarian sets up your team for success to ensure streamlined and error-free design, production, and sourcing. Register today for free.