Artificial Intelligence (AI) is buzzword of the day, with wild predictions about the replacement of engineers for writing code, testing, and performing actual designs based on constraints. The growth is highly accelerated by the money being poured into building the technology. Goldman Sachs says that it is expected that the global investment in AI could exceed $10 T over the next decade. (Thanks, AI for finding this reference for me!) The Stargate AI initiative alone is expected to be funded at $500 B over the next four years. (Thanks again, AI). That’s quite a lot of money, which provides a clear reason that everyone in the engineering world is so highly focused. 

AI is still in its infancy, but in this blog, I will address how well AI supports very simple requests for engineering design support in power integrity and signal integrity applications. I will perform several targeted experiments, summarizing what I found AI did well and where it was lacking. I will also discuss my own opinion on whether it is a revolutionary tool or an overhyped buzzword.

Initial Thoughts

AI has been in widespread use for years in minor ways. For example, social media learns about our interests and behaviors and suggests content and products we may be interested in. Other examples include software such as Shazam, which can identify a song by listening to a short snippet. AI scours the internet to learn and form databases of information while also building connections from the data. This does not always go well. 

I was at the IEEE APEC conference in 2025 when I first learned that according to Google, I had (supposedly) passed away July 8, 2014. I had no idea! I immediately asked friends and family if it was true. I was relatively certain I hadn’t, but never having passed away before, how could I be sure? 

Spoiler alert: I am not dead, and the false information was corrected by Google within a few hours. The incorrect report was likely due to my association with someone of the same name. A quick internet search showed that the erroneous response was due to AI connecting me with a different Steve Sandler, an American handball player who did pass away on July 8, 2014. Perhaps this is one reason to be skeptical of AI; creating errors such as these on a larger scale is a scary thought.

Experiments

I want to be open minded, and I want to perform a much more scientific assessment. To that end, I defined several AI prompts for this article, each one related to engineering with a focus on SI or PI. Some requests are generative, meaning that I aski AI to create something, and in other cases, I ask for education or relevant information. I used both Google AI Mode and Copilot, since these are readily available to everyone at no cost. Other AI programs might do better. After evaluating the responses, I will provide my own thoughts and opinions.

Example 1: Provide a List of Substitutes for AVX 1.2 nH Inductor

Prompt: Find an equivalent to this AVX chip inductor L06031R2BGSTR.

This seems to be a simple request. We currently use a 1.2 nH chip inductor to verify our impedance measurement setup. It’s the tightest tolerance (+/-8%) chip inductor in a hand solderable size. It is also more expensive than many others. While this is not an issue, it does offer a simple opportunity.

The AI response summarized the specifications of the AVX part, provided reasonable disclaimers, and suggested I use the search tool on a distributor website, such as Digikey. It did provide a few manufacturers series to consider.

Example 2: Differential Amplifier 

Prompt: Design a differential amplifier with a 80 dB DC common-mode rejection ratio (CMRR) and 100 MHz bandwidth.

The response offered a general response indicating that this was a very challenging requirement and suggested two op-amps worth considering. The first was a Texas Instrument THS4551, and the second was an Analog Devices MAX4445. The first is reasonable, but I failed to indicate that it needed to operate from +/-12 V and be single ended. The second is a 2X gain amplifier, but only 55 dB CMRR. I tried refining the prompt.

Prompt: Select a single ended op-amp for a 100 MHz differential amplifier with >100 MHz bandwidth, +/-12V operation, flat response, and 2X gain

This second prompt suggested an ADA4899-1, which has sufficient bandwidth and CMRR, but won’t operate from +/-12 V. Additional recommendations are to use 0.1% resistors, which won’t come close to the required 90 dB CMRR. To be fair, this prompt is quite a challenge, so it is not surprising that a solution was not identified.

Example 3: Thin Film Resistor Trimming and Effect on High Frequency Performance

Prompt: Thin Film Resistor Trimming and its Effect on High Frequency Performance

This request provided a very reasonable overview explaining the impacts and a variety of choices.

AI Response:

Impact of Trimming on High Frequency Performance:

While laser trimming is essential for achieving desired resistance values, it can also introduce or affect parasitic elements like inductance and capacitance, which become more significant at high frequencies. 

• Parasitic Impedance: The trimming process can cause a roll-off of the resistor's impedance at frequencies exceeding 100 MHz. This is due to internal capacitance from pad coupling and external capacitance from coupling to the board's ground plane, as well as internal and external inductance related to the resistor's physical dimensions and the skin effect.
• Trim Style Matters: The specific trim pattern used significantly affects the resistor's impedance performance. For example, the balanced edge sense trim method is often preferred for high-frequency thin film resistors, exhibiting better impedance performance than L- or S-cut trims.
• Optimized Trimming for High Frequencies: To minimize parasitic effects and maintain performance at high frequencies, optimized laser trimming techniques are used, especially for high-frequency applications like those extending up to 67 GHz.
• Termination Style Impact: The way the resistor is terminated (e.g., wrap-around vs. flip-chip) also affects high-frequency performance. Flip-chip termination is often preferred at higher frequencies because it minimizes inductive losses.
• Board Effects: The test board and mounting style can also influence high-frequency performance. For instance, a microstrip style with a ground plane on the backside can create different dielectric constants between the resistor and ground, impacting RF performance. 

In summary, while laser trimming is a valuable tool for adjusting thin film resistor values, particularly for precision and high-frequency applications, its impact on the resistor's parasitic impedance, influenced by trim style and termination method, must be carefully considered and optimized for high-frequency performance. 

Example 4: Learn about Power Integrity

Prompt: Recommend publications to get acquainted with power integrity quickly

This prompt did provide some good recommendations. I am pleased to be included in the recommendations, along with Eric Bogatin and Raj Nair. Notably missing is Istvan Novak, who is probably the godfather of power integrity and one of the most prolific power integrity engineers I know.

AI Response: 

To quickly get acquainted with power integrity (PI), here are some recommended publications and resources that cover essential concepts and practical applications:

Books:

• Signal and Power Integrity – Simplified by Eric Bogatin: This widely recommended book provides a solid foundation in both signal and power integrity, simplifying complex concepts for easy understanding.
• Principles of Power Integrity for PDN Design–Simplified by Eric Bogatin and Larry Smith: This book specifically focuses on power integrity for power distribution network (PDN) design, offering a practical approach for robust and cost-effective designs in high-speed digital products.
• Power Integrity Analysis and Management for Integrated Circuits by Rajendran Nair and Donald Bennett: This book provides a foundational understanding of power integrity, including practical aspects of power delivery and chip design optimization.
• Power Integrity Using ADS by Steven M. Sandler and Anto K. Davis: This resource offers a quick and easy guide to power integrity design practices and simulation techniques using Keysight's PathWave Advanced Design System (ADS).
• Power Integrity: The Faraday Press Edition by Steven M. Sandler: Another recommended book by Steven Sandler, focusing on proven techniques for achieving high fidelity measurements in power integrity applications. 

Online Resources:

• Signal Integrity Academy by Teledyne LeCroy: This online platform provides a wealth of technical content on signal and power integrity, including courses, webinars, and podcasts.
• Power Integrity Boot Camp for Designers Seminar by Keysight: This seminar offers practical guidance on power integrity design, including hands-on simulations and measurements.
• Four Power Integrity Webinars Worth Your Time: This article on the Signal Integrity Journal highlights key takeaways from webinars on topics like decoupling capacitor optimization, power integrity measurements, and SI/PI challenges in high-performance computing.
• Power Integrity How-to Video Series by Keysight: This series provides practical videos on various aspects of power integrity design, such as selecting VRMs, optimizing decoupling capacitors, and identifying power delivery noise problems.
• Power Integrity Analysis in Your PCB Design Software by Altium: This resource discusses integrated power integrity analysis capabilities within Altium Designer, emphasizing the use of simulation tools for identifying and addressing issues.
• What is Power Integrity? Understanding PDNs by Signal Edge Solutions: This article provides a good introduction to power integrity, covering definitions, PDNs, and the importance of including all system elements in simulations.
• What is Power Integrity and How Do I Measure It? by Keysight Blogs: This blog post explains what power integrity is, what causes issues, and how to measure it, including using oscilloscopes and DVMs. 

Courses:

• Master the Fundamentals of Power Integrity and Power Supply Testing by FEDEVEL: This online course offers hands-on lab demonstrations and classroom-style lectures to help you master power supply measurements and power integrity testing. 

These resources help to gain a strong understanding of power integrity concepts, design considerations, analysis techniques, and measurement practices to ensure reliable and efficient power delivery in electronic systems. Power integrity is crucial for successful signal integrity and addressing both is essential for modern electronic designs. 

Example 5: Create a Table of Publications with Descriptions 

Prompt: Create a chronological list of all Picotest publications including books, articles, presentations, and webinars.

To be honest, this request was because I would really like to create a searchable database of my publications, and this seemed like a fast and inexpensive way to achieve that goal. In the end, the prompt returned only a small fraction of my publications; however, it did a decent job providing information for those that it found and included in the list. 

AI Response:

Here is a chronological list of Picotest authored publications, primarily by Steve Sandler, formatted as a table with brief descriptions:

(For a continuation of the AI-provided table, please find refer to page 2.)

Note: This table provides a chronological overview based on available information. Some exact publication dates may require further research. The "Ongoing" entries represent resources and activities that are continually being updated and added to.

AI responses may include mistakes. Learn more

Courses | Picotest Online

MASTER THE FUNDAMENTALS OF POWER INTEGRITY AND POWER SUPPLY TESTING * WHAT WILL YOU LEARN? How to test Power Distribution Systems and Power Supplies. * HOW WILL...

Example 6: Suggest a Click-Bait Style Title for an Article

Prompt: I need a click bait title for your voltage measurements for power integrity are fatally flawed

I didn’t choose any of the suggested titles, but to be honest, it did help me brainstorm. 

AI Response: Sure! Here are a few attention-grabbing titles you might like:

• The Shocking Truth: Why Your Voltage Measurements Are Leading to Disaster!
• Fatal Flaws in Power Integrity: Are Your Voltage Readings Lying to You?
• Voltage Measurement Myths: The Silent Killer of Power Integrity!
• Your Voltage Readings Are Failing You—Here’s the Hidden Danger!

Direct & Shocking:

• Your Voltage Measurements for Power Integrity: FATALLY FLAWED!
• The Shocking Truth: Your Power Integrity Voltage Measurements Are Lying to You!
• EXPOSED: Why Your Power Integrity Voltage Measurements Are Worthless!
• Are Your Voltage Measurements FATALLY FLAWED? Power Integrity Secrets Revealed!

Curiosity-Driven:

• You Won't Believe What's Wrong with Your Power Integrity Voltage Measurements!
• Unlock the Secrets: Why Your Power Integrity Voltage Measurements Are Failing You!
• Stop Making This Mistake: Your Power Integrity Voltage Measurements Are Flawed!
• The Hidden Danger in Your Power Integrity Voltage Measurements!

Benefit-Oriented:

• Finally: Get Accurate Voltage Measurements for Power Integrity (Avoid These Fatal Flaws!)
• Fix Your Power Integrity NOW: Learn Why Your Voltage Measurements Are Wrong!
• Master Power Integrity: Overcome the Fatal Flaws in Your Voltage Measurements!

Remember to ensure the content delivers on the promise of the title! While clickbait can increase engagement, it's important to provide valuable information and solutions to the problems discussed.

Conclusion

My simple experiments provided mixed and limited results. I appreciated the notes stating that the response could include errors as well as the often-included commentary on the challenges and level of difficulty required to achieve my request. But how did AI do? That’s the question, right?

My own opinion, based on these examples, is that AI isn’t quite helpful in the generative space, at least not yet. While my differential amplifier request was a difficult one, some of the recommendations, such as op-amps and resistor tolerances, were just wrong. The search for a replacement part was less effective than a search on a distributor site, but to be fair, the AI suggested using the distributor site. In the request for a list of Picotest authored publications, AI provided a very short list, though it did get them in chronological order and provided a short summary. In broader, less generative prompts, such as finding ways to learn about power integrity or how trimming impacts high frequency performance of resistors, it did provide solid, though basic, insight.

In areas of editing grammar, producing text, and summarizing information, it did reasonably well. In response to a prompt to edit this article, it did a reasonable job extracting and editing the writing, but failed to include any of the AI generated responses. I suppose this is still helpful, but falls quite short of miraculous.

Continued AI Response for Example 5: