Breaking News
Qualcomm Preps IoT, IBM Upgrades Watson
5/16/2014

< Previous   Image 2 of 2   

Michael Rhodin, head of IBM's Watson group, reached out to big data entrepreneurs at TIE Con.
Michael Rhodin, head of IBM's Watson group, reached out to big data entrepreneurs at TIE Con.

< Previous   Image 2 of 2   

Return to Article

View Comments: Newest First | Oldest First | Threaded View
Anand.Yaligar
User Rank
Author
Re : Qualcomm Preps IoT, IBM Upgrades Watson
Anand.Yaligar   5/23/2014 7:42:57 AM
NO RATINGS
The position where right now the IOT stands in the global market has been summarized aptly in the form of 3 sentences spoken by Steve Mollenkopf. Chip making for the IOT is slow at the moment and QUALCOMM CEO understands it. However the direction where his company will go after IOT starts playing in the electronics supply chain will be governed by what business policies his company has adopted in the advent of IOT.

docdivakar
User Rank
Author
Re: Qualcomm Preps IoT
docdivakar   5/20/2014 12:55:26 PM
NO RATINGS
Rick, while I realize TiECon is not the venue to go much more in-depth, I wish the Qualcomm exec had disclosed some info on the planned development of I-o-T Silicon. It seems to me that he could have addressed areas where Qualcomm intends make a distinction that the I-o-T devices need, such as:

1. Power management including enegy harvesting
2. More sensor integration with CMOS processes
3. Low power multi-band / multi-protocol radios, etc.

MP Divakar

alex_m1
User Rank
Author
Re: Can you guess
alex_m1   5/19/2014 9:53:14 AM
NO RATINGS
Dr, those kinds of things might be an industry effort. And you can always start by creating a small set of examles that are aimed at what engineers think are "corner cases" and let developers work with that, it might take a long time to conquer those. Also defining some of those test cases might be done by non experts with minimal training, because humans are good at language.

 

 

There are also some kinds of tricks to gather examples, like google started with books which have been translated as data for theil translation algorithm.And  this is different than ASIC, in ASIC everything needs to work, here only some cases.

But yes, it's challenging and takes a long time.

 

 

DrFPGA
User Rank
Author
Re: Can you guess
DrFPGA   5/19/2014 9:33:28 AM
NO RATINGS
"You test this by creating a huge number of example sentences and their representations and testing how close watson got it."

Yep- the need to create a large number of example sentences to cover all of your application seems like a testing nightmare to me. It's like creating a large number of test packets to exercise your ethernet implementation. Takes years...

alex_m1
User Rank
Author
Re: Can you guess
alex_m1   5/19/2014 5:08:34 AM
NO RATINGS
Dr, Watson is composed on many components each with each own testing and debugging methods. For example there's a part that break sentences to a data structure that represents meaning. You test this by creating a huge number of example sentences and their representations and testing how close watson got it.


Debugging : some parts of watson use machine learning. Machine learning themselfs are math algorithms so you debug them  at that level. But than if you setup the parameters incorrectly and you see weird results , you try another combination of parameters ,probably guided by machine learning theory.

I guess similar methods scale to the top levels of watson .

If you're interested in more details, there are some guys from watson on quora , usually answr questions.

DrFPGA
User Rank
Author
Can you guess
DrFPGA   5/18/2014 3:41:27 PM
NO RATINGS
what a testbench will look like to prove out a design on Watson? I can't...

Anyone know how to test and debug?

Most Recent Comments
michigan0
 
SteveHarris0
 
realjjj
 
SteveHarris0
 
SteveHarris0
 
VicVat
 
Les_Slater
 
SSDWEM
 
witeken
Most Recent Messages
9/25/2016
4:48:30 PM
michigan0 Sang Kim First, 28nm bulk is in volume manufacturing for several years by the major semiconductor companies but not 28nm FDSOI today yet. Why not? Simply because unlike 28nm bulk the LDD(Lightly Doped Drain) to minimize hot carrier generation can't be implemented in 28nm FDSOI. Furthermore, hot carrier reliability becomes worse with scaling, That is the major reason why 28nm FDSOI is not manufacturable today and will not be. Second, how can you suppress the leakage currents from such ultra short 7nm due to the short channel effects? How thin SOI thickness is required to prevent punch-through of un-dopped 7nm FDSOI? Possibly less than 4nm. Depositing such an ultra thin film less then 4nm filum uniformly and reliably over 12" wafers at the manufacturing line is extremely difficult or not even manufacturable. If not manufacturable, the 7nm FDSOI debate is over!Third, what happens when hot carriers are generated near the drain at normal operation of 7nm FDSOI? Electrons go to the positively biased drain with no harm but where the holes to go? The holes can't go to the substrate because of the thin BOX layer. Some holes may become trapped at the BOX layer causing Vt shift. However, the vast majority of holes drift through the the un-dopped SOI channel toward the N+Source,...

Datasheets.com Parts Search

185 million searchable parts
(please enter a part number or hit search to begin)
Like Us on Facebook
EE Times on Twitter
EE Times Twitter Feed