The Opportunities of Electronics Machine Learning

Key Takeaways

  • Learn the basics of machine learning and its process.

  • Compare and contrast the uses of machine learning and deep learning.

  • A real-world example of how electronic machine learning has improved chip design.

Half-brain lobe, half via fanout

Few topics receive as much buzz in technical circles as artificial intelligence. As computing power grows and control systems better integrate the wealth of data available at a given moment, automated systems are making decisions in real-time with a complexity approaching that of human thought. No longer seen as mere number-crunchers, electronics are rapidly being employed to handle tasks that at one time would have seemed fit for science fiction. More and more, electronics machine learning is finding inroads into both consumer and business technology as a method of further increasing the speed and efficiency of tasks that may have once been deemed too abstract for computers to process.

But, what is machine learning, exactly? How can it intelligently guide electronics to better decision-making and efficiency? To start, we will consider the core operating guidelines of machine learning, its role in lightweight power designs, and finally, one of its many current-day uses in assisting chip designers as high-end technology brushes up against the limits of the physical world.

Machine Learning: Turning Electronics Into Studious Pupils

Machine learning is a subset of artificial intelligence whereby a system is trained to recognize data by looking for patterns, features, or characteristics that otherwise discern it from other data that could potentially be mistaken for it. This sorting process, known as data cleaning, helps teach the algorithm to determine which category or categories to tag a new datum entry with by using some detection method, such as edges or boundaries, and comparing it against known category entries. A brief overview of the steps to training and improving the reliability of the machine vision (or another form of recognition):

  • Data collection of the inputs into the algorithm that will serve as a teaching tool. Traditionally, these items are common image sources e.g., fruit or animals, but industry usage could instead visually assess PCB assemblies for misplaced components, tombstoning, or other assembly defects at a rate faster than human visual processing.

  • Data cleaning combs through data looking for incomplete or incorrect data that may serve to confuse the learning algorithm.

  • Training, modeling, and refining the algorithm by manually selecting the features under consideration, looking at the predictive power of the model with new data sets, and further increasing the efficacy of the algorithm based upon those results.

The power of electronics machine learning is that the heuristic model can be extended far beyond computer vision; analysis of machinery and factory equipment can study qualities like sound, vibration, and much more to predictively analyze when equipment needs to be repaired or replaced. The models are only limited by the abilities of the sensors themselves–any data that can be quantified can be fed to a machine-learning algorithm to build a data-driven (and eventually self-sufficient) model. 

Machine Learning’s Low-Power Advantage Over Deep Learning

Machine learning and deep learning are both subsets of artificial intelligence, but the approach to processing differs immensely, which has significant implications as to how and where it is deployed. Machine learning uses some human guidance to develop its vision, such as using colors and contrast to search for edges or other identifying features. Deep learning, on the other hand, bypasses human interaction and uses a more robust and voluminous dataset to build its predictive function. The tradeoff for deep learning is that the data must be much better cleaned before being fed to the algorithm, as there is no human guidance assisting in distinguishing factors. In general, deep learning can be thought to be a more predictive model that takes more time, data, and computing power to reach its greatest potential

 low-power or power-conscious designs.

The relative computational increase in deep learning models inhibits the adoption of low-power or power-conscious designs. When appropriate, lightweight power options will always favor a lower draw over a more comprehensive modeling effort. While individual sensors will draw nominal power alone, buildings looking to reduce energy usage may find some additional cost savings with a machine learning approach instead of deep learning.

 Timeline showing adoption of artificial intelligence, machine learning, and deep learning

The hierarchy of artificial intelligence, machine learning, and deep learning

Folding Intelligent Design Into Workflows With Electronics Machine Learning

Machine learning is a tool available to any dataset, provided sensor technology allows for tracking of relevant parameters needed for feedback. On its own, the advantages machine learning offers in handling large amounts of data provide endless opportunities for industrial IoT. Perhaps more interesting to the chip designers of the world is what machine learning can offer to designers.  Specifically, as the demand for chip processes continues its steady march towards the picometer scale, it becomes ever more difficult to balance the power and spacing needs of new chipsets. Optimizations can take teams of engineers significant working hours to meet the complexities of industry needs and wants.

Here, machine learning can also lend a hand to ease the design process. Engineers can create an incentive-based algorithm for whatever section of design they wish to optimize. Computing power can be aggregated to solve intricate design problems logically without requiring constant human oversight. Leaving the extreme fine-tuning to powerful computing also frees up designers and allows them to take a role more akin to oversight while realizing tangible improvements in speed and power consumption. What’s more, multiple optimization efforts can continue in parallel based on any user-defined optimization setting. Assuming the requisite technology and computing ability, design optimizations can be easily forked and reintegrated to even more expeditiously finalize chipsets.

When it comes time to incorporate electronics machine learning into your designs or workspaces, Cadence’s PCB design and analysis software offer powerful tools to simplify design implementations without sacrificing any depth of control. Leading electronics providers rely on Cadence products to optimize power, space, and energy needs for a wide variety of market applications. If you’re looking to learn more about our innovative solutions, talk to our team of experts or subscribe to our YouTube channel.

About the Author

Cadence PCB solutions is a complete front to back design tool to enable fast and efficient product creation. Cadence enables users accurately shorten design cycles to hand off to manufacturing through modern, IPC-2581 industry standard.

Follow on Linkedin Visit Website More Content by Cadence PCB Solutions
Previous Article
Disadvantages of 8-Bit Microcontrollers
Disadvantages of 8-Bit Microcontrollers

Disadvantages of the 8-bit microcontroller include memory size constraints and speed when compared to highe...

Next Article
An Introduction to Electronic Component Naming Conventions
An Introduction to Electronic Component Naming Conventions

Learn why it is so critical that designers understand and adhere to electronic component naming conventions.

OrCAD Free Trial

Try OrCAD Today