IEEE

Postscript to Nov 5 Early History of Silicon Valley Panel Session: Moore’s Law

Notices:

  • If  you’d like to join/leave our Event Notification email list, please send email requesting same to: brianberg@gmail.com.  To be removed from the list, please send email to Brian with Subject line=SV History UNSUBSCRIBE
  • If you have any comments, issues, suggestions, feedback related to our committee or this website, please email:  aweissberger@sbcglobal.net
  • Please check http://ithistory.org/blog for related tech history blogs by Ted Hoff, Gordon Moore and Alan Weissberger
  • The blog post below is a replica of email sent to our email list with selected reader comments appended.

Appreciation:

I hope those in attendance (97 total) enjoyed yesterday’s (Nov 5th) panel as much as I did. Paul, Ted and Norm did a superb job of telling many stories of what is now known as “Silicon Valley.”

Moore’s Law was not conceived at Intel:

In my closing remarks I noted that Moore’s law was responsible for almost all the advances in electronics over the last forty years.  Most people know that, but think that Moore’s law was conceived at Intel.  That’s wrong!  It was postulated in 1965, when Gordon Moore was still at Fairchild Semiconductor as R&D Director.  The “law” applied to both bipolar and MOS chips at the time.  Here are two references with embedded hyperlinks:
…………………………………………………………………………………………………………….
While Fairchild did the ground breaking work on MOS LSI in the mid to late 1960s and all through the 70s, they were not successful in commercializing the technology (see rebuttal from CHM’s David Laws below).  
Most folks don’t even know Fairchild had an MOS development effort!   Fairchild had other secrets too.  Did you know that Fairchild Systems Technology was working on a MOS/LSI micro-controller for the Sentry IC Testers in 1969-1970?  
And that they were designing 8/16/31 bit minicomputers called “Sprint” from March to Sept of 1970 (when the department was shut down)?  [This author was one of two CPU designers, working on a microcode assembler and simulator.]
 
Another key point is that Intel transformed Moore’s law from a theory to reality in the early 1970s, starting with the 1103 1K DRAM.  The volume produced enabled the MOS process to scale to higher chip densities (# transistors per chip) such that Intel could develop custom chips (e.g. for Busicom) and later microprocessors such as the 4004, 8008 and 8080.
 
In the early and mid 1970s, no one believed that MOS LSI processors would ever be used as the CPU for computers- bipolar LSI bit slices were the solution for that.  Wrong again…as MOS circuit densities, speeds and (lower) power consumption all combined such that MOS overtook bipolar LSI and created the PC industry.  
 
From my summary of a recent Commonwealth Club event:  Author Michael Malone at the Commonwealth Club: The Story Behind Intel
“The most important thing Malone said during his talk (including the Q&A session) was that Intel was the “keeper of Moore’s law,” which has been responsible for almost all the advances in electronics for several decades.  That’s due to Intel being able to continue to  advance the state of the art in semiconductor processing and manufacturing which enables them to pack more transistors on a given die size, increase speed, and reduce power consumption.”
………………………………………………………………………………………………………
 
Bottom line is that MOS digital circuits have followed Moore’s law for over 40 years.  Much later, flash memory did too.  But not analog ICs or bipolar LSIs.
………………………………………………………
Comment from Paul Zander:
  • Moore’s Law started as an empirical observation.  It become entrenched as a business mandate by Intel and other companies.
  • In the early 1900’s various companies were rushing to use the new Marconi wireless to replace telegraph and telephone wires.
  • By the end of the 1900’s various companies were replacing the airwaves with cables for TV.   The word, wireless, has been applied to eliminating cables between nearby boxes, e.g. Bluetooth, Wi-Fi, etc.

………………………………………………………………………………………..

Contrary opinion from Tom Gardner:

For those interested in Moore’s Law might want to take a look at, “The Lives and Death of Moore’s Law” According to this author  (Ilkka Tuomi from Finland) as of 2002:

“Contrary to popular claims, it appears that the common versions of Moore’s Law have not been valid during the last decades.  As semiconductors are becoming important in economy and society, Moore’s Law is now becoming an increasingly misleading predictor of future developments.”

If anything it has become less relevant in the subsequent decade.

…………………………………………………………………………………………..

Comments from David Laws:

Fairchild not successful at commercializing MOS LSIs:  Under Corrigan (CEO from 1975-79) Fairchild was quite successful in commercializing MOS technology, they shipped hundreds of millions of $ in revenue. His problem was the lack of successful proprietary products to generate sustainable profits. 

Fairchild annual reports for the 10 years from 1970 to 79 show total semiconductor sales of $2.6B. MOS comprised approximately $300M of that number. Products included 4K DRAMs, 8K EPROMs, and the F8, 3870 and 6800 microprocessors. According to a 1978 Dataquest report, Fairchild shipped 655 million units of the F8. Intel shipped 510M units of the 8080, and Motorola 435M units of the 6800. The company also had a digital watch division (they built models for Tiffany and others as well as their own brand) and a consumer video game business that used MOS devices. The latter’s major claim to fame was the introduction of the ROM-based game cartridge that was quickly adopted by Atari.

“Most folks don’t even know Fairchild had an MOS development effort!” – True. Even fewer know that Fairchild invented CMOS! See:http://www.computerhistory.org/semiconductor/timeline/1963-CMOS.html

…………………………………………………………………………..

Comment from Ted Hoff, PhD- employee #12 at Intel in 1968:

Gordon always considered his work more of an observation than a law. It did provide an important guideline for determining the optimum complexity of ICs at a given point in time. In the early days of Intel, somewhere around 10% yield was about optimum, and applying Moore’s law helped meet that goal. Assuming 100 die sites per wafer, and a $50.00 processed wafer cost, each die would cost $5.00. Allowing $1.00 to package the die would result in a manufacturing cost of $6.00.  Sell it in quantity for $10.00 to $20.00 and you make a nice profit.

Double the size of the die, and you have less than 50 die sites, and yields around 1%. Each chip would cost more like $100 so you would need to charge more like $200-$300 for the same profit, and would likely be competing with SSI/MSI designs which were assumed to be on a PC board at $1.00 to $2.00 per IC. The higher price would probably discourage applications, so production volumes would be lower and there would be less improvement based on the learning curve.

Given Moore’s law, just wait a few years, and that more complex chip would be very manufacturable. In the past few years I have heard stories of perfect wafers–huge wafers of extremely complex chips running at 100% yield.

You can go the other way as well, assume half the die size and over 200 sites per wafer with abut 30% yield. Now the dice cost more like 80 cents each but if packaging still costs $1.00, the final cost is close to $2.00. It probably comes pretty close to the original total cost for the customer, but based on Moore’s law, it would soon become obsolete.

The advances in IC technology have helped reduce the cost of computing enormously, but there are still many areas where I think more computing progress might have been made–e.g. natural language processing, language translation, security, reliability, etc. Just because computer technology advances does not automatically help computer usage in certain areas.

Somewhere, someone needs to devote a lot of time working to solve those applications. Regarding IC design, the profits reaped by the semiconductor industry helped to motivate the IC industry to develop ever more capable tools and those tools helped reduce engineering costs. Standardization also helps reduce microprocessor design engineering cost, but helps in increasing applications for them. The effect is to move the engineering burden from microprocessor chip design to software and firmware development.

Alan, you were correct in noting that Moore’s original observation was made when he was at Fairchild. He considered that it applied to both MOS and bipolar designs. At Intel, most of the progress was made in MOS technology, although the Schottky bipolar design was a significant step in allowing bigger/more complex bipolar chips.

………………………………………………………………………………………………………………………………..

Nov 5th Presentation slides Thanks to Paul Wesling for assembling this slide deck.
  • Nov 5th Videos (you can watch the entire video or individual video segments/”snippets” with captions). Thanks to Ken Pyle for creating these videos and Paul Wesling for reviewing and approving)

    ………………………………………………………………………………………………………………………..

 

Our next technical meeting is Dec 2nd: A conversation with Stanford’s Leslie Berlin and Henry Lowood will be at TI auditorium in Santa Clara, CA starting at 6pm with a networking reception/light dinner.

Dec 2 Meeting: Perspective from Stanford’s Silicon Valley Archives

 

Leave a Reply