Software Defined Instruments
SDI- System-on-Chip with Compute-Centric Architecture will end reign of established T&M
"It is Time for instrument manufacturers to make the computer the center of their universe", was the beginning sentence of a paper by Rick Robinson, Agilent, in EETimes on June 5, 00, unfortunately, what follows only scratches the surface. This "technical" development is foreseen today as a certainty. (E.g., James Truchard, Carsten Thomsen, National Instruments, Pat Gelsinger, Intel, and others.) And the established T&M industry is also aware of this. However, they are not aware or acknowledge that the architecture and circuits for these "Compute-Centric" instruments will also be of a new kind, they will be SOFTWARE DEFINED, which will have the effect of totally "disrupting" their present "Business Model". In the past, this has always resulted in the demise of the affected industry, and all indications are that history will repeat itself. (See Prof. Christensen: The Innovator's Dilemma, When New Technologies Cause Great Firms to Fail. And also a "superb" recent paper by Girish Mhatre in the EETimes Supplement, Sept.002, titled THE ART of CHANGE, specifically targeting electronic developments which will have this effect during the coming years. (A tool-box for identifying "disruptive-technologies")
One reason these great changes are happening for T&M is a paradigm shift as part of the information age. Electronic Design Automation, EDA, and the PC-workstation are already the primary design tools for all of electronic engineering. Instruments are now used mainly for the final design phase, the verification, and for maintenance. And because of the complexity of today's designs, only the "simulation and signal-integrity programs" know the shape and timing of waveforms and their test-point locations. (For waveforms, timing, schematic and board layout with test points --- see the shapely <7 of 9>) This requires instruments to be controlled by EDA or maintenance software and be tightly coupled to the computer. The Interchangeable-Virtual-Instrument Foundation, IVI, IEEE-group is now providing the programming standards.
What the T&M industry is presently providing for some instruments, and planning to do more of in the future, is to equip their "old" designs with computer interfaces such as USB1/2, IEEE 1394, and the "legacy" IEEE 488 ports. Will it work? Yes, if you are willing to add to your $3000 Workstation a $30,000 DSO-peripheral, which you use only for the last 10% of your design, such as the verification of your latest OC-12 or OC-48 or OC-192, systems. (You have just chosen a Luxury Ferrari as your peripheral, Congratulation!) The excessive prices for high-performance instruments, stand-alone or PC interfaced, are caused largely by the T&M industries determination to ignore Moore's Law. If congress ever adopt this law, they could be indicted for fraud! Here is the other major reason why <disruptive> changes are coming for the T&M Industry. As T. J. Rogers, CEO of Cypress, so vividly expressed it: If you don't stay on the Moore's Law curve with everybody else, you're going to get hammered to a pulp.
Submicron IC technology has arrived at a point in time when whole instruments can be design for "Instrument On a Chip" (IOC) implementation. The counter argument of the T&M industry is that their type of market will not justify costly IOC development. Dan Strassberg, long time T&M editor of EDN magazine, writes:
Then there is the economics of test and measurement (T&M). The T&M industry manufactures thousands of highly specialized products, few of which achieve annual volumes that exceed 10,000 units. These sales volumes are minuscule when compared with those of such products as PC peripherals. Yet, the development costs can be similar. For T&M manufacturers to recover those costs, a product must remain in production for years.
Very true! - And here are the underlying reasons for it: The "mass-market" of PCs and peripherals is made possible by their general-purpose nature, which allows all specialized tasks to be performed in <software>; this generates one value-system. Today's instrument designs are specialized in <hardware>, which generates a very "fragmented-market" and a different value-system, not supporting SoC development. (If products of both types are manufactured and designed in one firm, such as the original HP, it creates a "clash of value systems".)
The Road to SDI, --- SOFTWARE DEFINED INSTRUMENTS
Knowing the reason why present instruments can't take advantage of costly SoC designs, which is the latest achievements of "Moore's Law", also points to the solutions. Create SoC modules performing test-instrument functions of a basic or generic kind. (Motorola's digital DNA slogan.) Combine these new "software defined" modules into Compute-Centric Instrument systems, which will then perform a multitude of specialized task in <software>, just as PCs and their Peripherals do. Fragmentation of the market is significantly reduced and the volume will rise to the point where it becomes economical to develop these generic IoC instrument modules.
A simple solution says the business mind, ----- Impossible! Shouts the old "legacy" engineer. Well, I never said it was simple, --- For the first of these Software Defined Instruments I chose the FLAGSHIP of the T&M industry, the GigaHertz DSO, and put it on a PCMCIA-card for a cost less than 1/20 of what the industry is charging. It took one patent, an international design contest win for the architecture, and a total of 5 engineering-man-years spread-out over 20 years. No new "million gate" submicron technology is used, instead I used a "heritage SoC"; -- you ask what that is? These are the microcomputer chips of old, such as the Motorola MC68HC11 and others. If new software/hardware architecture can perform a large part of the system functions on this one chip, you basically have a SoC.
Gorden Moore notes that up to and including the
.25-um generation of processes, changes in device speed and density could be
attributed to: two-thirds part processing and one-part creativity
[architecture]. In the future, the ratios will undoubtedly reverse.
------------------------------------ Tets Maniwa, Editor, isd magazine
Rick Rashid, director of Microsoft's research division, calls it "Moore's Law on Vacation", because its job is being done by new architecture instead. He was referring to architectural advances in 3D graphic boards exceeding Moore's Law by an order of magnitude.
While this first "Software Defined DSO" is based on new architecture, the circuitry is eminently suited for SoC implementation for nearly all-peripheral chips surrounding the CPU core, and that with conventional process technology. The total system would consist of one (or two) SoCs. Other improvements could address the low MIPS rating of the heritage Motorola chip, which actually limits the speed of the DSO response-time, a better choice would have been an ARM-core. A complete high-end GHz DSO as a generic SoC module might appear to be an over-kill for specialized applications, such as embedded waveform monitoring on other systems. But since 85% of this design is software, reducing it results in no cost savings. Software Defined Instruments will always have more capabilities than needed for any specialized task.
For "Price versus Bandwidth" of present products, Go BW-Cost-1 - For cost ratio of lowest to highest Bandwidth for the new design, GO edge-1 - For Reference designs GO-Reference.
To show what I mean by saying that "PC-peripherals have a general purpose nature which allows all specialized task to be performed in <software>", let me give a short review of the desktop printer evolution. Starting with the ancient Teletype we progressed to IBM typewriters with a proprietary computer interface (a mechanical wonder!). Then came the mechanical pin dot-matrix printer with different fonts and primitive black-and-white graphics. If you wanted to print color graphics or even photos you had to use an expensive specialized printer. Then came the laser printer, which gave us back the high quality and added different fonts and graphics, but size and cost did not make them a desktop peripheral. Today we have the latest generations of Inkjet printers which are truly "general purpose print engines". Printing sharp fonts, high resolution graphics for desktop and for fax service; and these wonderful photo prints with vibrant colors, and the latest Epson printers now have a 50 year life on photo prints, better than color film! --- Is this really all done by software? You bet! My 133 MHz Pentium grinds to a crawl when I am printing one of my favorite fall-color photos on my $150 HP-712C DeskJet.
If you now expect me to explain how the thousands of highly specialized products of the T&M firms can be reduced to a handful of universal SoCs, you mistake me for Santa Claus. But let me make one of my Grandiose Statements*. Given the advances in SoC clock speed and performance at GigaHertz frequencies, which Moore's Law predicts for, let's say the year 2005, many "specialized" instruments could be designed with just two universal SoC modules. These would provide just two functions for all "electromagnetic" tests; one a high precision digitizer (DSO), the other a high resolution ARB (Arbitrary function generator), both operating in the very high multiple GigaHertz range. Both chips are Software Defined Instrument Blocks, where <programs> running on multi GIPS PC-processors will handle all <specific> test-functions. --- Anyone who watched the TV shows StarTreck Voyager knows that the sensor-array of the star-ship (her instruments) can be reconfigured, by just telling the computer to do it --- nobody ever runs around the ship doing hardware changes. ---------------
* Actually, the architectural concepts I am describing here are nothing new, credits must go to James Truchard, CEO of National Instruments. He again wrote about it in Electronic Design, July 24, 00, Broad-Based Technologies Are Revolutionizing Instrumentation, which (in a different way) makes the same "technical" statement. ----- The issue he is not addressing in this article is that a "software defined", compute-centric architecture also opens the door to economical SoC development, which will then have fatal market consequences for all T&M firms, including the hardware products of NI. (See <Disruptive Innovation 101> and Christensen's book part one, chapter 4.) But just as it was for other major discoveries, e.g. Relativity, Quantum mechanics, etc., it will take decades before Christensen's discoveries are fully understood and applied. And while the first "atom bomb" made people believe in Relativity, it will take several catastrophic failures of major industries before the leaders of this industry realize that "Christensen's Criteria", together with "Moore's Law" will predict the future of all electronic business for very many years to comes.
To download the pictures for the first of the future Software Defined Instruments. GO >Tricorder2G ---Is it a DSO? - Is it a TDR? - Is it a 4-1/2 digit Voltmeter? - Is it a Spectrum analyzer? - Is it a Display for Schematics and Layouts from your EDA program? - Is it all of this? - YES! -(It's "Software Defined"- stupid, - it's Superman)- But also your trusted PDA for all your personal stuff. - Or a wonderful excuse to get a PDA! - And how about the first economy version of this design? GO > Tricorder200 - How about "The other" PC-Card DSO on the market? Want to construct your own Instrument? Look at Instrument Platform reference designs. ---- To see what these new designs will do to the "Handheld Instrument Market", go: OPPORTUNITY --- Back to Index
"The Software is the Instrument"
(Or how I ended up with a "Software Defined Instrument", just to save hardware cost)
Most engineers I talked to believe that this line from James Truchard, CEO and Founder of National Instruments, is just a metaphor. --- Surprise, it's not! The novel architecture of this DSO has evolved by transforming more and more of the functions of a DSO into software. Originally to save hardware cost, which, of course, is a well-known concept of embedded system designs, but then something else happened. The architecture became extremely flexible, because the functioning of the system blocks was now predominantly "software-defined". (See: Equivalency for Hardware/Software)
Initially, my incentives were all the peripheral circuits included by Motorola into their HC11/12 type processor chips. E.g., like other chips, the HC11/12 has 8 channels for A/D conversion, but I had use for only two --- the low frequency real-time digitizers. Working on an early breadboard, I discovered that I could rapidly get a stable trigger by hanging a voltmeter on the output of the trigger circuit and adjusting the input-level for a symmetric output-swing. Here was a use for another two A/D channels detecting amplitude and approximate trigger frequency.
To get the necessary range of input attenuation for this DSO, the scope probes must have a 1 to 10 switch. If the "program" is not being told the switch setting, the V/Div display will be wrong. Remembering how the cheap game-controllers (joystick) for PCs work, I am now using a second resistor and switch contact in the probe and an A/D channel to get the indication. But the PC game-controllers are using "one" 8 bit A/D to sense up to 32 various switches --- and oscilloscopes can use many different probes. You might get the drift; the "program" can now distinguish between a dozen different probes. (Current probes, high-voltage, FET probes, optical probes, etc.) Trivial? Yes, but if we don't count the cost of the resistor, it's for free, and the "other" scope manufacturers either don't provide this, or charge you a mint. This kind of "serendipity" repeated itself in other parts of the system. When I added a D/A converter to the time-base for the Zoom feature (a ten-time expansion of the trace at a cursor point, see picture on first page.), I found that the same circuit could now also do the "Anti Aliasing" feature --- the time-base was truly "software defined"!
The real interesting part of this "Software Defined" design started when I began to use a software algorithm for the digitizing process itself and for the time-base operation. (With the ancient HC11/12, I needed to drop into machine code, but still could use more speed.) The flexibility that this design-step provides is incredible! I still have not yet discovered all advantages and uses of it. Any kind of digitizing algorithm can be implemented; and the time-base can go forwards, backwards or sideways. It can jump over areas or zero-in on them with high-resolution sampling.
For a long time, for instance, I was "stuck", believing that the concept of DIS needed a byte of RAM for every sample displayed on the scope screen. Then it came to me that I could do long traces for PC type computers with less RAM by dividing these into two or more groups; such as odd and even, with only minor change in the time-base algorithm. I had been <thinking-inside-the-box> all these years, but with the concept of "Software-Defined" design the box became much wider. Will we ever get outside the box? No, we will just be thinking in bigger boxes. ------ Back to Index