Last week, while watching a video of my friend Dave Jones tear down an old Fluke 91 ScopeMeter DSO (digital sampling oscilloscope), I suddenly realized that this 20-year-old piece of test gear brought the decade-long decline of ASIC (application-specific integrated circuit) design starts into sharp relief.
Dave cracked open his eBay ScopeMeter and revealed the instrument's two main circuit boards: one digital and one analog. The ScopeMeter's digital board contains two ASICs plus an 83C196 mask-programmed, 16-bit microcontroller from Intel's old MCS-96 family. The SoC (system-on-chip) era started in 1995, and Fluke built this DSO in 1994, just before processors started to climb aboard ASICs, which is why there's a separate on-board microcontroller.
The digital ASICs on the ScopeMeter's digital board appear to be a Flash-memory/keyboard controller (labeled "Fluke MASIC," fabricated by Hitachi) and a timebase/trigger/event-counter/display processor (labeled "Fluke D-ASIC," fabricated by LSI Corp.). There's a third ASIC fabricated by Philips -- an analog chip on the DSO's analog board -- that incorporates amplifiers, analog switches, and track-and-hold circuits for two analog signal channels. That's two digital ASICs and one analog ASIC designed into a relatively small instrument that sold for $1,000 or $2,000 nearly 20 years ago.
Here's a screen capture taken from Dave's EEVBlog video showing the ScopeMeter's digital board on the left. I've circled the two digital ASICs and the Intel 83C196 microcontroller in red:
Now, it's true that the Fluke ScopeMeters sold in relatively high volumes for a piece of test equipment, but it's hard to believe that any such instrument designed today would incorporate two digital ASICs and a microcontroller. In fact, it would not. Had Fluke designed this ScopeMeter a few years later, chances are good that both digital ASICs and the mask-programmed microcontroller would have all wound up being designed into one SoC -- probably an ARM microprocessor core.
This consolidation of multiple digital ASICs and processor chips into one device is one of the significant trends that has caused the number of ASIC design starts to fall dramatically since the mid-1990s. According to published EDA (Electronic Design Automation) research reports, the number of ASIC design starts peaked at roughly 11,000 per year in 1997 -- then came the triple threat of:
- Increasing consolidation of multiple ASICs, processors, and other devices into one SoC as permitted by Moore's-Law scaling. (The Fluke 91 ScopeMeter is clearly a perfect candidate for this kind of ASIC consolidation.)
- The dotcom bust of 1999/2000 followed by the 2008/2009 recession.
- Rapidly escalating design and NRE (non-recurring engineering) costs for leading-edge nanometer SoCs.
As a result of these three trends, the number of ASIC design starts seems to have permanently fallen by about 80 percent from the 1997 peak through today.
To Page 2