Podcast: Play in new window | Download
Subscribe: Apple Podcasts | RSS
Welcome Jerry Twomey (Effective Electrons) author of the book, Applied Embedded Electronics: Design Essentials for Robust Systems. Chris first heard Jerry on Embedded.fm last year.
- Jerry’s Background and Book Motivation: Jerry shares his quick history, moving from the Boston area to San Jose (Silicon Valley) and eventually to San Diego, where he has worked across diverse sectors including consumer electronics, aerospace, defense projects, DARPA research, and medical electronics. His book focuses on how to develop robust systems, providing guidance that is timeless rather than applications manuals that quickly become outdated.
- The Analog Problem: Although modern systems may be digital end-to-end, Jerry emphasizes that the predominant causes of failure and design difficulties are often analog in nature. Academic study often teaches ideal signals but neglects real-world issues like inductance, noise, and cross-coupling.
- Consulting Experience & Troubleshooting: Jerry discusses being called in to fix systems that failed strenuous regulatory testing for medical devices, where reliability is first and foremost (similar to an aerospace way of thinking). Failures often stemmed from basic issues like a lack of ESD protection, absence of error correction in data streams, insufficient detection of errors, and common mode noise rejection problems.
- High-Speed Data and Signal Integrity: At high data rates, communication becomes a “communications channel problem,” not truly a digital one. When bits are underneath a tenth of a nanosecond, the communication turns into multiple standing wave transitions. The two primary limits on performance are rise and fall times and distance traveled.
- Real-World Applications: Jerry has worked extensively on medical devices, including early-generation Dexcom glucose monitoring systems (two on-body monitors and a hospital insulin pump/monitor), and a wearable EEG monitor. He also worked on a system that required packing five video cameras into an endoscope distal head, measuring 11 mm in diameter and 13 mm long.
- Architecting Systems and Identifying Bottlenecks: When starting a new project, Jerry suggests defining needs and interfaces and looking at the system as a black box. Engineering time should focus on the bottleneck—the hardest part of the system. For medical implantables, this might be minimizing power consumption down to virtually nothing, which could take up 90% of the effort.
- Power System Design: Jerry advises purchasing commercial AC-to-DC converters due to competitive pricing. He notes that switching supplies (buck converters) commonly introduce noise that can lead to EMI failures or corrupt sensitive analog front ends. A classic case of “digital thinking in an analog scenario” is when a sensitive analog front end is powered by a noisy switching converter.
- Working with Embedded Teams: Jerry prefers guiding embedded teams toward “self-discovery,” using bench time and empirical measurement (such as comparing grounds on a scope) to demonstrate non-ideal connections and grounding issues. He advises against the “seagull manager” approach.
- Grounding Best Practices: For integrated circuits (chips), designs must be fully differential because securing a good hard ground reference is impossible. On singular circuit boards, a common uncut ground plane (dedicated ground plane, often multiple layers stitched together with vias) is the recommended approach. Cutting the ground plane is discouraged as it can create a slot antenna, increasing the signal radiating from the board by about 7 dB. Jerry has published rules on grounding.
- Engineering Intuition vs. LLMs: Jerry notes that intuition is gathered through painful learning experiences and guidance from experienced designers. He expresses concern over the reliance on LLMs (Language Learning Models), which, while improving, can confidently provide incorrect answers, especially regarding complex topics like signal grounding.
- Limits to Moore’s Law: CMOS scaling is approaching physical limits, likely unable to go below 10 or 11 nanometers. Modern performance gains are achieved through more parallel processing, not significantly faster clock rates, which have plateaued around 5 GHz due to parasitics and timing limitations. Jerry’s article discusses this topic.
- RISC Architectures: The industry benefits from migrating to RISC (Reduced Instruction Set Computing) architectures (like ARM) because they eliminate useless architecture and transistors associated with complex instruction sets (like x86).
Find Jerry on Effective Electrons and on LinkedIn
Leave a Reply