I was immediately struck by first figure of the first (introduction/motivation) chapter, which is a plot based on a 1998 paper by Hans Moravec called When Will Computer Hardware Match the Human Brain? Here is the plot, overlaid with an additional point from the present world-champion supercomputer, the 1.5 peta FLOPS (that’s 1.5×10¹⁵ floating-point operations per second), $133 million, custom machine Roadrunner that IBM built for Los Alamos National Lab:

(Strictly speaking, Moravec’s paper has instructions/s but Roadrunner is spec’d in FLOPS, but given the three-decade-per-tick log scale this difference is negligible.)

Some observations:

• If you don’t have $100 million to spend on a supercomputer, just wait ~20 years and you’ll get the same power on your desktop for a few thousand dollars.
• Far from slowing down, it seems the pace of improvement is accelerating. Maybe it’s because each generation of computers enables the next by allowing more and more sophisticated engineering using said faster computers and the EDA and CAD tools that run on them?
• According to Moravec’s criterion, the world smartest thing isn’t a human (with a mere 0.1 peta instructions/s): Roadrunner is fifteen times "smarter." Presently, fifteen humans are a lot cheaper than one $133 million Roadrunner (just its electricity bill is ~$4M/year), but if a $1,330 desktop computer in 2030 is smarter than its user, why bother with the user or their desk?

Is Roadrunner really as smart as a human? Will the trend continue or flatten off? Will we get a one peta FLOPS machine on our desktops by 2030? Will it still take Windows four minutes to reboot? What do you think? Post a comment below!

Acknowledgement: Thanks to Michael Mirmak who kindly alerted me about Hall & Heck’s book shortly after I posted some book reviews. Yikes! That was back in June. How time flies…

If you enjoyed this post, please spread the word by sharing and bookmarking:

• Timing Analysis and Simulation for Signal Integrity Engineers by Greg Edlund
  This book takes a different approach to signal integrity then the "traditional" Bogatin and Johnson & Graham approach (see below) in that the starting point is not the interconnect, but that diminutive prima donna of signal integrity, the flip-flop in the receiving chip. In my mind’s eye I imagine this scene: "If I don’t see a clean zero or one during my set up and hold periods," lectures the flip-flop, one hand on hip, the other wagging a threatening finger, "I’m going to go metastable." And so the game begins. How to deliver on that demand? Greg’s twenty years – and counting – experience as a hands-on signal integrity engineer makes his textbook authoritative and his narrative style makes the book accessible.
  One minor issue I do have with this book is the issues raised on page 222: “A short survey of introductory physics textbooks uncovers an unsettling fact: it does not seem possible to define magnetic field in general terms using more fundamental quantities.”
  Not so. The magnetic component of force between two particles is well defined by the Biot-Savart law and the B field can be defined in such a way as to intermediate between cause (B₁ = μ₀q₁v₁ x r /(4πr³)) and effect (the magnetic component of the Lorentz force F₂₁ = q₂v₂ x B₁).
• A Signal Integrity Engineer’s Companion Real-Time Test and Measurement and Design Simulation (Geoff Lawday, David Ireland, and Greg Edlund)
  Some of the chapters (2, 4, 9) in this books are the same (give or take a revision or two) as ones in the book above, so this book should be considered if the non-overlapping material is of value to you. Not sure why the publisher "double-dips" in this way.
• Signal Integrity – Simplified (2ed due for release July 27, 2009) by Eric Bogatin
  It was Eric who coined the phrases "There are two kinds of engineers — those who have signal integrity problems, and those who will." and (with respect to all signal integrity issues) "It depends."
  I haven’t seen the second edition yet, but the first edition has the "traditional" signal integrity focus on the properties of the interconnect, the impairments that it inflicts on the poor signal, and how to design around them within the limits imposed by physical laws. Eric also has a new book coming out in September – Signal Integrity Characterization Techniques – co-authored with my colleague Mike Resso. I’ll review it when I can.
  Update July 29, 2009
  I received the 2ed of the now renamed Signal and Power Integrity -Simplified. Eric has revised the original eleven chapters, a net addition of about 6 pages of material on topics like differential pairs and losses that have become more prominent since the first edition in 2004. But the major change is the addition of two new chapters, one on S-parameters and one on power distribution networks.
• High Speed Digital Design: A Handbook of Black Magic and High Speed Signal Propagation: Advanced Black Magic by Howard Johnson and Martin Graham
  Published in 1993 and 2003 respectively, one frustration with these two very popular books is that I can never quite decided whether the latter is volume two of the former, or whether it’s the second edition. There’s certainly a lot of overlap between the two, but the second isn’t a true superset of the first. It smacks of another case of Prentice Hall "double-dipping." Both volumes/editions have a "traditional" focus on the PRBS signal in the frequency- and time-domains, and its interaction with the interconnect. Personally, I prefer the flow of Eric’s book for "cover-to-cover" reading, but there’s a lot of good reference material and anecdotes here that I dip into when needed.

Other texts that are on my ever expanding to do list. I’ll try to add reviews here someday? Or you can add your thoughts in the box below.

• The SPICE Book by Andrei Vladimirescu
• Jitter, Noise, and Signal Integrity at High-Speed by Mike Peng Li
• Learn Signal Integrity Design Principles With Mathcad: Interactive textbook/CD for EMC/EMI/PI/SI by David Norte
• Semiconductor Modeling For Simulating Signal, Power, and Electromagnetic Integrity by Roy G. Leventhal and Lynne Green
• Power Integrity Modeling and Design for Semiconductors and Systems by Madhavan Swaminathan and A. Ege Engin
• High-Speed Digital System Design: A Handbook of Interconnect Theory and Design Practices by Stephen H. Hall, Garrett W. Hall, and James A. McCall
• Advanced Signal Integrity for High-Speed Digital Designs by Stephen H. Hall and Howard L. Heck

If you enjoyed this post, please spread the word by sharing and bookmarking:

Michael Malone says this of his book, Bill & Dave: How Hewlett and Packard Built the World’s Greatest Company:

“This is not [just] a history of the Hewlett-Packard Company, or a book of business theory, or a definitive biography of William Hewlett and David Packard. I have chosen to write this book this way because of the desperate need the business world has right now for an archetype of enlightened management, enduring quality, and perpetual innovation. It is not enough to simply tell the story of Hewlett, Packard and their company. What are needed are the why? and the how?”

So it’s not the (parochial) connection between HP and my company Agilent (an HP spin-out) but the why? and how? that makes this book relevant to signal integrity engineering and to this blog. Sometimes when you have a challenge, you have to dig deep. What can we learn about their four decades of growth, in an industry where most companies fail, or have a success as ephemeral as the reverberation of the word “Netscape” around the floor of the New York Stock Exchange?

The Malone quote above identifies “enlightened management, enduring quality, and perpetual innovation” but it omits HP’s uncanny knack of identifying a need that can be fulfilled by innovation. HP didn’t sell a single computer until the 27th year of its existence (the room-sized 2116A in 1966). They got into the business because they heard that their customers had need to connect HP instruments and HP plotters into “smart” instruments that automated time-consuming, repetitive, and error-prone manual data collection, analysis, and visualization.

How about signal integrity? Where is the need that can be fulfilled by innovation there? A recent article in The Economist contains this graphic about data center spending:

Data centers are rapidly growing consumers of electricity and are expected to spend about $40 billion US dollars to consume more than 100 billion kilowatt-hours of electricity a year by 2011.

Although overall spending in power, cooling, and administration, is rising dramatically, the fraction of this going to hardware vendors is shrinking dramatically. The absolute amount is flat or declining, even as the number of servers and the quantity of compute power is increasing.

The message is clear: to get a larger “wallet share” of IT spending, hardware vendors must focus on two burning needs of their data center customers: lower cost of ownership via ever low power consumption, and ease of administration. Signal integrity folks can help differentiate their companies products with lower voltage (hence lower power) interconnects and chips (see for example the Spansion EcoRAM postioning) and by administrative innovations like hot plugging and redundancy.

I know that a lot has changed since Bill and Dave passed the reigns of HP to John Young in 1978. But don’t you agree there’s still something to learn from the HP Way? Please add your comments by clicking on the link below.

Book Review – Bill & Dave – MP3 version.

If you enjoyed this post, please spread the word by sharing and bookmarking: