More bad news for European electronics

[Bernard Murphy]

While Nokia slipped in phone handsets, it had seemed that Ericsson would continue to keep Europe relevant in communication infrastructure. Now even that seems in jeopardy.

Ericsson remains in the red, market outlook darkens | Reuters

 

[Eric Esteve]

This title seems to demonstrate a lack of understanding of the European Telecom market!

Nokia has been kicked out of the mobile (not a scoop in 2017), but the company has been completely re-engineered around the Networking business.
Nokia is doing so bad that the company has made €23.6 B in 2016, this represent +90% compared with 2015!
See below: http://www.nokia.com/sites/default/files/files/nokia_20f16_full_report_2_1.pdf (page 13)

Please take this type of information into account before dropping such title...

 

[Bernard Murphy]

I acknowledge that Nokia is perhaps doing better recently than I implied Eric (though analysts do not seem to believe they are out of the woods yet: NOK Stock: Should You Buy Nokia Oyj (ADR) (NOK) Stock? 3 Pros, 3 Cons | InvestorPlace), but Ericsson was always seen (at least from the US) as a lynch-pin in communications and is clearly struggling at present. Another example is ST which even the most ardent supporter would have to admit is not the company it once was. At least from a US-centric view it is difficult to see much to cheer (yet) among the big players in the European electronics market. ARM, Infineon, a few other (relatively smaller) players still look good but I don't yet see a return to the strong position Europe held say 10 years ago. I understand the desire to root for the home team (I'm European too, at least pre-Brexit). But turning this around starts with acknowledging the reality, no?

 

[Staf_Verhaegen]

I acknowledge that Nokia is perhaps doing better recently than I implied Eric (though analysts do not seem to believe they are out of the woods yet: NOK Stock: Should You Buy Nokia Oyj (ADR) (NOK) Stock? 3 Pros, 3 Cons | InvestorPlace), but Ericsson was always seen (at least from the US) as a lynch-pin in communications and is clearly struggling at present. Another example is ST which even the most ardent supporter would have to admit is not the company it once was. At least from a US-centric view it is difficult to see much to cheer (yet) among the big players in the European electronics market. ARM, Infineon, a few other (relatively smaller) players still look good but I don't yet see a return to the strong position Europe held say 10 years ago. I understand the desire to root for the home team (I'm European too, at least pre-Brexit). But turning this around starts with acknowledging the reality, no?

I think one of the biggest causes of all this is that until recently there was no talk about European microelectronics industry. You had the French/Italian industry with ST, the German microelectronics cluster around Dresden and the Ireland microelectronics cluster. For equipment you have the Dutch industry around Eindhoven (e.g. ASML) with big link to Leuven (e.g. IMEC). Only recently I have seen talk on European microelectronics with cooperation between German and French clusters. Unfortunately this realisation that they can't do it alone anymore may be too late.

 

[Eric Esteve]

If we talk about microelectronics industry, I tend to agree with Staff... but the recent re-positionning done by ST (Auto, Sensors, Power), on line with the CEA-LETI and Fraunhofer decision to invest in research in areas where the European industry is strong (FDSOI, Auto, Industrial IoT) demonstrate a realistic positionning.
Advanced foundry: mostly in Asia (Taiwan with TSMC, Korea with Samsung) + US (Intel Foundry, GlobalFoundry)
Wireless phone: IC processed in the above mentioned foundries, systems manufactured in China
Data center, HPC: US with Intel, GloFo, but not in Europe

This leave room to develop in certain growing segments (Auto, IoT, Industrial IoT, Medical,...) where Europe can be strong, even if not alone to compete. For example, Audi has just announced the launch of most advanced driverless car (July 11th), see the Financial times dated July 11...

 

[Staf_Verhaegen]

This leave room to develop in certain growing segments (Auto, IoT, Industrial IoT, Medical,...) where Europe can be strong, even if not alone to compete. For example, Audi has just announced the launch of most advanced driverless car (July 11th), see the Financial times dated July 11...

With names like XFab, Melexis, OnSemi, Bosch, Siemens, Philips, etc. I think Europe is already big in those areas. Problem is that it is mostly a boring, non-sexy B2B affair on mature nodes and revenue for microelectronics part is small compared to the 'smart'/wearable things; especially as the big Asian foundries also have automotive qualified processes.

 

[smeyer0028]

I think this thread is seeing European semiconductor industry incorrectly because everyone who reads Semiwiki seems to believe in AI PR hype. Namely, a singular event from deep learning in imminent because computers now learn from each other at silicon switching speed.

The real innovations in more useful computers is advances in sensors and actuators where European STMicroelectronics is the world leader. James Lighthill argued in 1972 in his "Lighthill Report" to the British government that all AI was doing was taking credit for advances in feed back engineering involving sensors and actuators. It is still valid. AI made the same claims back then that computers would soon (by 2000?) surpass human intelligence. Deep learning was called knowledge bases. Lighthill argued that intelligent algorithms ran into the combinatorial explosion (in modern term the search space has more points than pico-seconds the universe has existed). Also back then AI promised rapid increase in computer intelligence from computers competing with each other. He pointed out just as with modern deep learning programs where computers learn (from each other?), all that is happening is computer programs are searching for local extrema in very large search spaces.

European semiconductor development follows the Lighthill pattern focusing on control engineering rather than the impossible task of building superior intelligence (GPUs, hardware neural nets, quantum computers, communication for government surveillance etc.). It is very successful in that focused area.

 

[hist78]

European semiconductor development follows the Lighthill pattern focusing on control engineering rather than the impossible task of building superior intelligence (GPUs, hardware neural nets, quantum computers, communication for government surveillance etc.). It is very successful in that focused area.

Can you elaborate a little bit about why it's impossible? Thanks.

 

[count]

It's worth adding that for all the hype around AI in self driving cars, the most commonly used algorithm in the space is a variation of MPC which comes from traditional control engineering and was invented in the 1980s for controlling refineries.

 

[smeyer0028]

Can you elaborate a little bit about why it's impossible? Thanks.

Newton Professor of physics at Cambridge University explained combinatorial explosion this way in 1972. Modern computers that say are 10**9 faster than 1972 computers allow for exhaustive searching of larger problems than 1972 but even small problems require exhaustive searching of rather common 10*100 size search spaces so the 10**9 is a drop in the bucket.

See "Light Hill Report" at this URL:

Lighthill Report

Title is "Artificial Intelligence a General Survey"

Lighthill described combinatorial explosion this way in section 3 category A:

"Before going into such detail he is inclined, as a mathematician, to single out one rather general cause for the disappointments that have been experienced: failure to recognize the implications of the combinatorial explosion. This is a general obstacle to the construction of a self-organising system on a large knowledge base which results from the explosive growth of any combinatorial expression, representing numbers of possible ways of grouping elements of the knowledge base according to particular rules, as the base's size increases."

In 1972 AI argued that the combinatorial explosion can be avoided with heuristics. Heuristics were abandoned by AI. I think current AI argues the combinatorial explosion can be avoided by neural networks that by analogy (with no proof) work the same as human pattern recognition. My understanding of the current AI answer may be wrong.

I am not sure of examples because many assume Turing Machines are the correct model of computation. I believe the correct model is Von Neumann's MRAM machine (random access machine with unit multiply and a bounded number of unbounded size memory cells). One example is that it is easy to construct functions on the rational numbers that require exhaustive search of every rational number in the space to find an extreme point with a particular property.

 

[Staf_Verhaegen]

Newton Professor of physics at Cambridge University explained combinatorial explosion this way in 1972. Modern computers that say are 10**9 faster than 1972 computers allow for exhaustive searching of larger problems than 1972 but even small problems require exhaustive searching of rather common 10*100 size search spaces so the 10**9 is a drop in the bucket.

That's why you have heuristics and the heuristics only need to be better than the human ones to be more intelligent.

 

[count]

Feature selection in machine learning is an NP hard problem. There aren't any good heuristics for determining which features to examine, it takes domain expertise.