You are currently viewing SemiWiki as a guest which gives you limited access to the site. To view blog comments and experience other SemiWiki features you must be a registered member. Registration is fast, simple, and absolutely free so please, join our community today!

  • Steve Furber has found his million ARM cores

    Some people say that everything in our lives happens for a reason. As we wrote Part I of “Mobile Unleashed”, the origin story of ARM architecture and its main progenitors Steve Furber and Sophie Wilson, we found what seemed like an obvious technological breakthrough was far from an overnight success – and it led to fascinating twists and turns.

    Article: Intel Goes Vertical to Guarantee PC Growth-arm1-processor-evaluation-board.jpg

    An example is the point Furber and Wilson were working on a business plan for a stand-alone entity for ARM technology. One question loomed largest: How many units would sell? “I could never get the numbers to work. You have to sell millions before royalties start paying the bills. We couldn’t imagine selling millions of these things, let alone billions,” said Furber. Without a suitable answer, he decided to leave for a faculty position at the University of Manchester.

    As we know, ARM found massive success and went on to sell 60 billion processor cores and counting. How did things turn out for Steve Furber? His seminal text on ARM architecture guided an entire generation of chip designers. His early research converted ARM to an asynchronous design, AMULET, without much commercial success. His work still centers on ARM architecture.



    “Silicon Brain: 1,000,000 ARM cores” – Computerphile

    Since its launch in 2007, Furber has directed the SpiNNaker project – a loose amalgamation for Spiking Neural Network Architecture. In short, it seeks to emulate real-time brain operations. The planned hardware targets 1 million ARM cores packaged in ten 19” racks, equaling about 1% of the power of a human brain, or as Furber likes to say about 10 mouse brains. Each rack uses chips with 18 cores each, 48 chips on each board, then 24 boards stacked tightly 5 frames high – 103,860 cores total. The low power consumption of ARM again proves critical.

    The first 100,000 core rack is complete and running in Manchester. In the above video shot in August 2015, Furber shows a SpiNNaker board and its “hexagonal subsurface of a toroid” topology, then shows a lab with one of the frames, then the full 19” rack with its plethora of over-the-top Ethernet cabling. He then discusses some software concepts for mapping neurons, and bigger ideas surrounding human memory.

    In another interview on SpiNNaker with Justin Richards just published, Furber is asked if he has anything in his career that he would have done differently, an obvious allusion to his decision to leave ARM just prior to its formation. Furber’s inspiring response:

    “I don’t think so, there are no decisions in my career path that I particularly regret and I think the advice I give to people is roughly the advice I follow myself, which is to make decisions that keep the maximum number of doors open. So look for opportunities, but when there’s nothing obvious staring you in the face then think about what subject creates the most possibilities in the area you’re interested in. Maximize the number of doors.”

    Amen. We invite you to read our new book with the details of the ARM origin story in Part I, along with chapters on Apple, Samsung, and Qualcomm in Part II, in paperback or Kindle format:

    Mobile Unleashed: The Origin and Evolution of ARM Processors In Our Devices”

    It is fitting to see how the man who had a hard time imagining sales of a million ARM cores now has 100,000 ARM cores in one rack, on the way to 1 million cores in a research facility exploring the human brain. I think if you asked Steve Furber on the day he decided to leave Acorn for the University of Manchester, he would not have imagined where he would be now.

    More articles from Don...