In just two generations, we have lived through six different technological revolutions that have transformed our economy, our careers, and our quality of life, for better or for worse. Generative AI is the latest.
The electronics revolution began in 1947 with the demonstration of the first transistor. Electronics are now part of every product, in every application across every aspect of our daily lives. Just recently, Nvidia announced the volume production of a single processor with 28 billion transistors, with devices planned in the next generation approaching 90 billion transistors.
The first personal cell phone was demonstrated in 1973. Today, there are over 6.8 billion cell phones in use. We have come to take for granted having access to instant communication with anyone around the world and 24/7 access to every movie, document, and piece of information ever created.
Information processing is an emergent application of the electronics revolution that enables embedded intelligence everywhere, from the microcontrollers in every appliance to the fastest supercomputer, the Frontier computer system at the Oak Ridge National Laboratory, which has more than 8 million cores and is capable of more than one exaflops.
The combination of electronics, intelligence, and mechanical action is the driving force behind the revolution in mechatronics. This spans the range from tiny micro-electromechanical systems, such as MEMs-based robots that travel through bloodstreams, to the robots that assemble cars.
It may be too late to have as profound an impact as is hoped, but the revolution in electrification is accelerating. All devices that are mobile, from personal devices to cars and planes, are migrating to battery power. Those devices, which must be tethered to a local power source, are connecting to the grid, which is evolving to enable multiple power generation sources. We are in the early phases of the electrification revolution. Stay tuned to see the impact it will have on all of our lives.
The sixth technology revolution that is accelerating in maturity is artificial intelligence (AI). The origins of AI had their start with pioneers such as Norbert Wiener’s cybernetics in the 1950s, and Marvin Minski’s neural network, SNARF, in the 1960s. Today, neural network-based machine learning algorithms have permeated many applications, such as ordering at fast food restaurants and autonomous cars.
The latest twist in the AI revolution is generative AI. The public was recently introduced to generative AI with the release of Chat Generative Pre-Trained Transformer (ChatGPT). While some have referred to this type of program as a glorified word completion application, it is becoming apparent that it is much more.
To learn more about generative AI, I took the natural next step and asked ChatGPT to tell me more about generative AI and its future. This is its (or should I say his, or her?) response:
“Generative AI is a subset of artificial intelligence that uses learned patterns to produce new data, such as text, images, or music, closely resembling existing examples. The future of generative AI holds promise in diverse fields. Enhanced creativity, personalization, and problem-solving capabilities will transform content creation, art, and research.”
The combination of all these technological revolutions and AI will affect our industry in two ways: firstly, how the electronics revolution will evolve to enable evermore powerful AI-powered systems, and secondly, how the application of AI systems will accelerate the design of evermore advanced electronic systems. It is sort of the ultimate form of introspection and self-replication: AI systems will help design the next AI systems.
I recently had the privilege of moderating a panel on “Designing 224G PAM-4 Systems for Generative AI Architectures,” with four industry experts: Karl Bois from Nvidia, Lennin Patra from Marvell, Gus Panella from Molex, and Chris Kapuscinski from Molex. A recording can be viewed here.
Three themes emerged from this panel that may have an impact on how future 224G systems are implemented.
Generative AI systems are parallel processing driven. Problems are parsed into many smaller pieces, each processed in parallel. Coherency between each branch is important. There is less overhead available for resending packets due to transmission errors, requiring a lower bit error rate than more resilient systems.
The design challenges span not just signal integrity, but mechanical integrity, power, thermal management, software, and manufacturability. This may create new working relationships between all the engineers across various companies who must come together to build a successful product.
Using the traditional approach of interface specs between modules and interoperability may cause the design margins to shrink below zero. Every AI system may end up being a custom, proprietary system with its own set of tradeoffs and system level optimization that drives design decisions rather than attempting any sort of interoperability.
The six revolutions we are all engulfed in have had a profound impact on our daily life. Generative AI will affect the daily lives of engineers, both inside and outside of our work.