( ESNUG 588 Item 3 ) ---------------------------------------------- [09/12/19]

Subject: Mo Faisal on analog IP inside $50 billion worth of AI/ML chips

              DAC'19 Troublemakers Panel in Las Vegas, NV


   Cooley: Mo, all this talk about machine learning.  Why are you doing
           analog stuff? That's dull.  

       Mo: Ouch, dull? That kind of hurts, John.  (laughter) 

           As Joe Sawicki said, there are two sides to this.  There's
           machine learning and AI on the EDA side, and then there
           are the actual ML chips that we're designing.  

           I've seen all kinds of crazy numbers about the market size
           and opportunity there.  You know, one of the most recent
           one I saw was that the AI chip market is going to be close
           to $50 billion within the next 5 years; that's a huge
           opportunity for everybody in this room.

           I'm really excited about it.  As a matter of fact, [Movellus]
           actually has customers in the AI chip space, so we are in it.  

           And when you actually look at the AI chips, there's two ends
           to the spectrum.  

           First are the edge AI chips.  These are very close to the data
           collection or the sensor device.  And the goal of that chip is 
           just minimize the amount of data that needs to be shipped to
           the cloud.  So, power and size are really important.  

           And we are the only game in town that can actually squeeze every 
           ounce of power savings out of these analog blocks which are 
           digitally generated.  

           We've even delivered sub-microwatt PLL to one of our IoT 
           customers.  That's the kind of level of efficiency we can
           deliver.

           At the opposite end of the spectrum are the big honking high
           performance workhorse AI chips, which are all hundreds of 
           thousands of cores.  And by the way, I think some of them are 
           using RISC-V cores, which is really good for Naveed.  So, so
           we are very much part of this AI revolution -- just on the
           chip side.  

   Cooley: But you're saying farmers are part of the AI revolution because 
           they feed all the engineers making the AI Chips.  [laughter]

       Mo: Well, actually, that's a good question.  But what's happening
           with AI, especially at the edge, is there is a lot of process 
           customization happening as well.  

           For example, when an AI company wants to do an AI algorithm at
           the edge, your traditional digital algorithms don't work anymore
           ...they're very very inefficient.  

           So, people are exploring memory-based architectures, analog 
           compute, add-and-subtract charge instead of binary bits and
           numbers.

           What happens there is all of a sudden, these chips, which are
           going to be everywhere -- part of the 50 billion market -- are 
           going to be analog heavy.  They could be as much as 60% analog.  

   Cooley: Right.  

       Mo: Well guess what?  Somebody's has to figure out how to make
           analog a bit more efficient.  And that's us [Movellus].

           Movellus beats True Circuits PLL/DLL IPs as #9 "Best of 2018"

   Cooley: Good farmer feeding the people.  Good.

       Mo: Exactly.

   Cooley: Okay.

        ----    ----    ----    ----    ----    ----   ----

Related Articles:

    Joe Sawicki on ML, Calibre, Solido, VC funding, and heuristics
    Anirudh on each Machine Learning engineer is worth $10 million
    Mo Faisal on analog IP inside $50 billion worth of AI/ML chips

   !!!     "It's not a BUG,
  /o o\  /  it's a FEATURE!"
 (  >  )
  \ - / 
  _] [_     (jcooley 1991)