Semiwiki EDA Webinar 800x100

The Appeal of a Multi-Purpose DSP

The Appeal of a Multi-Purpose DSP
by Bernard Murphy on 07-26-2016 at 9:45 am

When you think of a DSP IP, you tend to think of very targeted applications – for baseband signal processing or audio or vision perhaps. Whatever the application, sometimes you want a solution optimally tuned to that need: best possible performance and power in the smallest possible footprint. These needs will continue, but there’s growing interest in more flexible solutions to address multiple signal processing objectives through common functions and to support evolving requirements.

Automotive and IoT markets in particular are driving this demand for flexibility. ADAS, infotainment and sensor fusion require multiple applications processing multiple data types, in floating point for codecs for example, in fixed point for other applications, supporting multiple word sizes, signed and unsigned. But systems development teams don’t have armies of DSP software developers ready to develop assembly code and floating point libraries per signal processing function as needs and standards change.

What’s more, there’s increasing pressure to easily port existing software to DSP functions with the expectation that the compiler and the platform will take care of optimizations such as vectorization. In some ways, these multi-use DSP applications are increasingly demanding use-models we routinely expect in general-purpose computing, while still expecting high-performance.

Cadence has developed the Tensilica Fusion G3 DSP specifically to address these needs. In switching to a multi-purpose platform, customers may be willing to accept some performance compromise, but not a lot. So Cadence has optimized the architecture to give best possible performance, along with flexibility. The G3 offers single and double precision floating point, along with fixed point and a range of word sizes. It has a finely-tuned high-performance architecture, balancing MAC, load/store and ALU functions. The Tensilica group has also added a set of specialized operations on top of the base Xtensa instruction set architecture, to support optimizations for specific applications.

The G3 provides its own DMA controller and supports multi-banking for memory, helping you to get data in and out as fast as possible And naturally, since you’re going to be using this for multiple purposes, it supports multi-core usage. Debug is supported through Extensa Explorer and the G3 also connects to CoreSight debug and trace.


For DSP software developers, Cadence claims best-in-industry auto-vectorization through the compiler and an extensive library of IIR, FIR, FFT, 1D and 2D transform, math, statistics and other functions. This means it should be easy to port C or Matlab code developed for other architectures and still get high performance, without needing to dive down into assembly code. (You can still go to the assembly level if you want, but it’s less likely you will run into that need.)

The G3 was developed in close partnership with a customer who recently taped out their first G3-based design. While this is obviously a new release, Cadence are also seeing interest from other customers, especially around radar (automotive) applications. They expect the G3 will find a home in a lot of applications looking for best if not bleeding-edge performance with reduced software development costs (and schedule) and a higher degree of future-proofing. In audio, they see demand in surround-sound and active noise/echo cancellation (both of which require floating point), fingerprint recognition, communications (requiring complex floating point), image processing (scatter/gather on 8-bit data) and radar (requiring both floating point and fixed point).

The value proposition is pretty obvious. As DSPs become the go-to solution for more functions in a design, the industry is demanding more cost-effective solutions that are easier to adopt, easier to maintain and easier to adapt as requirements change. Simplifying total development costs through a common platform, without significant compromise in performance, is an obvious way to get there. You can learn more about the Fusion G3 HERE.

More articles by Bernard…


SMART sensors with OTP memory for the IIoT

SMART sensors with OTP memory for the IIoT
by Don Dingee on 07-25-2016 at 4:00 pm

A few years back before IoT became the buzzword, the industrial automation community had already talking about “smart sensors” since the mid-1990s. The impetus for those discussions was IEEE 1451, a family of standards for adding intelligence and wireless communications to sensors so they could be incorporated into field networks. Continue reading “SMART sensors with OTP memory for the IIoT”


Coming Up Next: ARM IoT ASICs!

Coming Up Next: ARM IoT ASICs!
by Daniel Nenni on 07-25-2016 at 12:00 pm

The History of ASICs is well documented in our book “Fabless: The Transformation of the Semiconductor Industry” which illustrates the earliest forms of design start driven collaboration. The history of ARM is well documented in our book “Mobile Unleashed” which illustrates an entire company culture based on design start driven collaboration.

That brings us to where we are today with hundreds if not thousands of system companies cobbling together IoT solutions using off-the-shelf chips. Amazon Echo, Nest Dropcam, and a Skybell are examples in my home.

The next phase of this transformation is what we call the IoT ASIC. Yes, the Internet of Things is a very fragmented market but it is also ultra-competitive so you will not survive if you are cobbling together systems. Take Apple for example (Chapter #8 in Mobile Unleashed), they went from cobbler to ASIC to full blown fabless semiconductor powerhouse in order to control their competitive destiny.

The Multi-billion dollar question here is: Who is going to deliver the next big IoT thing? The answer of course is just about anybody thanks to ARM and ASIC providers like Open-Silicon. In fact, just last month ARM selected Open-Silicon to join the ARM® Approved Design Partner program in conjunction with the ARM DesignStart™ portal:

“The new ARM Approved Design Partner program enables a powerful and extensive network of global design houses,” said Chris Shore, training product manager, ARM. “Open-Silicon has a successful track record in custom SoC design and manufacturing services as well as ASIC projects, and it has made significant investments in its ARM-based product services roadmap. As a member of the program, Open-Silicon can now play a valuable role in helping to enable the easy and rapid development of new ARM-based devices.”

This program builds on the ARM DesignStart™ portal, which offers SoC designers free access to ARM Cortex®-M0 processor IP for design, simulation and prototyping with the option to buy a simplified and standardized $40,000 fast track license. The design and ASIC houses selected to join the ARM Approved Design Partner program will provide expert support during development and manufacturing. They are experienced in developing custom SoCs using ARM processor IP, and have successfully completed a stringent ARM auditing process to ensure they meet the highest quality standards.

“The new ARM Approved Design Partner program enables a powerful and extensive network of global design houses,” said Chris Shore, training product manager, ARM. “Open-Silicon has a successful track record in custom SoC design and manufacturing services as well as ASIC projects, and it has made significant investments in its ARM-based product services roadmap. As a member of the program, Open-Silicon can now play a valuable role in helping to enable the easy and rapid development of new ARM-based devices.”

Open-Silicon’s selection for the ARM Approved Design Partner program validates the company’s investments in its ARM TCoE (Technology Center of Excellence), established in 2011, and its recent Spec2Chip IoT ASIC Platform, which was developed for low risk and reduced schedule custom SoC development. This scalable platform is based on the ARM Cortex-M processor, TrustZone® CryptoCell hardware-accelerated security technology and ARM mbed™ SDK. This platform allows IoT ASIC designs to be evaluated at the system level.

“ARM and Open-Silicon share the same vision for simplifying the path for system developers to deploy IoT platforms,” said Vasan Karighattam, VP of engineering, Open-Silicon. “Through this collaboration, both companies are paving the road to IoT innovation by facilitating the development of highly-differentiated custom SoC designs.”

About Open-Silicon

Open-Silicon transforms ideas into system-optimized ASIC solutions within the time-to-market parameters desired by customers. The company enhances the value of customers’ products by innovating at every stage of design — architecture, logic, physical, system, software, and IP — and then continues to partner to deliver fully tested silicon and platforms. Open-Silicon applies an open business model that enables the company to uniquely choose best-in-industry IP, design methodologies, tools, software, packaging, manufacturing, and test capabilities. The company has partnered with over 150 companies ranging from large semiconductor and systems manufacturers to high-profile start-ups, and has successfully completed over 300 designs and shipped over 120 million ASICs to date. Privately-held, Open-Silicon employs over 250 people in Silicon Valley and around the world. www.open-silicon.com


Formally Crossing the Chasm

Formally Crossing the Chasm
by Bernard Murphy on 07-25-2016 at 7:00 am

Formal verification for hardware was stuck for a long time with a reputation of being interesting but difficult to use and consequently limited to niche applications. Jasper worked hard to change this, particularly with their apps for JasperGold and I have been seeing more anecdotal information that mainstream adoption is growing. So I thought it would be interesting to ask Pete Hardee (marketing and product management for Jasper) what has changed in the industry and why.

Cadence now treats formal as one of the 4 legs of their verification strategy. They arguably have the market-leading solution in Jasper, but they wouldn’t make it a top-level component if the demand wasn’t there, so what’s different? According to Pete, virtually all of the top semis doing RTL design are now using formal, as are a lot of the fast-growing companies. And formal usage is growing within these companies. Adoption alone would suggest it’s no longer a niche application.

The reason for this change all comes down to coverage. Full dynamic SoC coverage is already well out of reach (because of size, complexity, 3[SUP]rd[/SUP]-party IP, software, …), but you still have to have high confidence by signoff. So verification engineers look for different ways to build confidence.

One way is through connectivity checks – separate questions of whether the IPs function and communicate correctly from whether you have connected them together correctly. Can I prove that all the IP in the design are hooked up per a specification I am willing to provide (usually a connectivity spreadsheet)? If you can completely prove this aspect of the design is correct, you are able to signoff a whole class of functional checks more completely than would ever be possible in simulation. This makes formal checks a natural approach when they’re sufficiently simple to use. Apps make them simple and that is growing adoption in verification teams.

A different class of problem is proving certain things cannot happen – something essentially impossible to prove in simulation for any reasonable-size design. A good example is proving that an encryption key cannot leak out to an insecure IP (or an IP being used in insecure mode), equally that it can’t be overwritten and that it remains secure even in the presence of faults. This isn’t an area where “reasonably” confident is an acceptable signoff, so you have to use formal methods.

Power management is a nightmare for coverage because you take an already massive mission-mode state space and exponentially expand it in switching between all the possible power variants. You can gain some confidence through dynamic verification, but complete proof that there are no gotchas in switching again requires formal, in this case supported by an app.

Pete also noted that fear of assertions and constraints seems to be on the decline. Solutions to certain properties you feel you must cover can’t always be pre-packaged in apps. This used to be when you’d ship the problem off to your team of verification PhDs. Now not so much apparently. Pete guesses that some verification teams bit the bullet in training (and maybe a little coercion), engineers in hot companies aren’t afraid of formal and real expertise in this area is looking increasingly valuable on a resume. “That stuff’s too hard” doesn’t seem to be something you want to be heard saying anymore.

Getting to high coverage near the end of the verification cycle is another driver. We all know that the last mile in coverage is really hard. Maybe that’s because a lot of the uncovered cases are unreachable. Proving that is a a real time-saver – you know if you formally can’t reach a state, you can safely drop it from coverage. And if the app proves you can reach it, it will provide you with an example that will help you build a test. Reachability analysis is an exploding area in formal because getting to maximum coverage is must-have.

Unsurprisingly, safety is driving more interest in formal, since safety is another area where “reasonable” coverage is not an acceptable goal. ISO26262 demands traceability of requirements, fitting well with formal which has well documented properties and constraints. In fault analysis, formal helps both in efficiency (why test a fault if it can’t be observed?) and in completeness (maybe a given fault-sim didn’t propagate a fault to an output or a checker, but would that always be the case?). Demonstrating safety to ASIL-D requirements in ISO26262 is again a must-have – expect that automotive safety will drive more growth for formal in multiple areas.


Pete added that he’s also seeing growth in exploring design state-space for bug-hunting. This is an interesting domain where deep state-space bugs can be missed by constrained-random and you can’t conclusively catch the bug with direct formal if the bug is too deep. JasperGold has engines which support a concept they call “elastic bounded model checking”, letting you do a guided search progressively deeper into state space while skipping states you don’t feel are of interest. One user group paper reported multiple critical bugs found at 100-400 cycles deep and one case at nearly 3000 cycles deep, far beyond reasonable bounds for conventional model checking.

Hopefully if you stayed with me this far you would have to agree that formal (especially in JasperGold) is covering a lot of bases. It’s no longer a niche application for specialists. It really has become a primary pillar of verification. I found a really useful way to understand more about how JasperGold is being used is to check out papers from the user group (JUG) conferences. You can get to papers from the last conference HERE (you’ll need to have an account with Cadence.com).

More articles by Bernard…


Car Theft Making a Comeback

Car Theft Making a Comeback
by Roger C. Lanctot on 07-24-2016 at 4:00 pm

In the U.K., where vehicle theft has been in a steep decline for the past 20 years, the most widespread advice given by police to car owners is: keep your car keys in your freezer. The most common source of vulnerability these days is the interception of RF signals between keyfobs and cars. For a time, several years ago, there was a rash of thefts that derived from the car owner’s inclination to leave car keys near their front doors.


The issue is timely as LoJack reminds us that July is National Vehicle Theft Protection Month. The company released an infographic to highlight its concerns: http://tinyurl.com/hxca92y

The proliferation of remote keyless entry and telematics is setting the stage for a renaissance in vehicle theft. Thieves are still interested in auto parts ranging from tires and catalytic converters to airbags, but grabbing the entire vehicle may be getting easier in some circles with the aid of code grabbers intercepting signals from keyfobs.

In the U.S., vehicle theft has also declined, though not as steeply as in the U.K. The decline has been steep enough to make life difficult for stolen vehicle tracking and recovery companies like LoJack. Data from the FBI for the first half of 2015 suggests a significant uptick in vehicle theft in the U.S. and the U.K. has also seen a recent spike.

Some recent thefts of FCA Jeep vehicles in Texas suggests that hacking may be taking the place of smash and grab style thefts of and from vehicles. The Texas report points to hacking thanks to video captured by a home owner of the car thief entering the vehicle and apparently using a laptop computer to start the car: http://tinyurl.com/j7pe7lm

The thefts in Texas are interesting and important from several perspectives. According to the news report, the police believe the thief tapped into the car’s on-board computer via the OBDII port and created his own key. FCA executives expressed their “concerns that the thieves may have gotten hold of a system used by dealers to pair the vehicles with a new key, one they already had in hand. That could be as simple as access to a dealer website where knowing a vehicle’s VIN, or unique identification number, can provide the necessary codes to marry car and key.”

Automotive Cybersecurity
The automotive industry has been wrestling with the issue of cybersecurity ever since IOActive analysts Chris Vlasek and Charlie Miller hacked their way into a Toyota Prius and a Ford Escape two years ago. The findings from these analysts, presented at a Black Hat cybersecurity event, was that cars are now frequently equipped with both telematics systems and automated parking systems – a combination that makes taking control of the vehicle locally or remotely both fun and potentially profitable.

The National Highway Traffic Safety Administration (NHTSA) got involved after Vlasek and Miller followed up their Toyota/Ford exploit (which involved a lot of dashboard disassembly) with the now-famous or infamous Jeep hack. The IOActive pair, who now work for Uber, demonstrated how they could remotely control the hacked Jeep – to the horror of FCA executives, regulators and Jeep owners.

Of course, the Jeep hack required some significant preparation and was not achieved without time spent reverse engineering code and penetrating the vehicle’s limited security preparations. In fact, the Jeep hack exposed a significant vulnerability which led to FCA initiating a recall and sending out USB software updates to owners of the effected vehicles.

Even after the Jeep hack, though, industry executives scratched their heads over why hackers would bother to hack cars. Up until recently car makers were content with their “security by obscurity” approach – ie. cars were just difficult enough to hack to make it not worth the effort.

But the prospect of vehicle theft combined with increasingly obvious security shortcomings may signal a turning point in the vehicle theft business. The latest data from the U.K.: vehicle theft is up 9.9%.

Software Updates
The Jeep hack exposed the seriousness of vehicle vulnerability and the extent to which car companies are ill prepared to respond. Vlasek and Miller’s hack was intended as a wake-up call to FCA and the industry – but their methods pushed the boundaries of ethical hacking.

Ethical hackers, like Lab Mouse, seek to penetrate a broad range of consumer products in the interest of finding and fixing flaws in security systems. Once a vulnerability is found the effected company is notified only after which are the details of the vulnerability published.

Vlasek and Miller revealed that certain Jeep vehicles lacked a necessary firewall between the infotainment system and the vehicle’s safety and powertrain systems. This created a big problem for FCA. Like most car makers, with the possible exception of Tesla, FCA is vulnerable to the sieve-like recall system in the U.S. where car makers struggle to find current vehicle owners – and vehicle owners ignore recall messages from their dealers and the car companies.

It is entirely possible that the hacker/thieves in Texas are exploiting the same vulnerability identified by Vlasek and Miller and taking advantage of the likelihood that the software-related recalls on effected models have not been seen to. We won’t actually know until the thieves are caught or stopped.

The thefts highlight the importance of over-the-air software update technology of the type used by Tesla Motors to add features and make code corrections in its Model S vehicles. FCA mailed out thumb drives with software updates – an approach widely frowned upon in the cybersecurity industry.

Dealers
There is yet another source of anxiety emanating from the Texas thefts. Dealers remain a weak link in the security chain. FCA’s suggestion that the Texas hacker/thieves might be accessing a dealer Website to clone keys is but one potential source of vulnerability. Disgruntled dealer employees have been known to wreak havoc with vehicle security and telematics systems.

Dealers are also a source of poor security hygiene because of the entire industry’s blasé attitude toward recall work. At the recent national gathering of automobile dealers, incoming NADA Director Jeff Carlson ridiculed the recall system, suggesting that most recalls did not represent urgent safety issues, based on industry research conducted by the auto makers.

In the FCA instance, the missing firewall is a vehicle theft waiting to happen. Vlasek and Miller may have had fun taking control, remotely, of a Jeep – but the real issue is theft.

The Texas Jeep thefts point up the greatest threat of weak vehicle cybersecurity: the return of widespread vehicle theft as a challenge for law enforcement and car owners. There has been a lot of fear-mongering around identity theft, vehicle ransom and remote control terror – but maybe we’re missing the most obvious threat in a world of connected cars – simple theft.


After the fatal Tesla crash, I still feel safe in my self-driving car

After the fatal Tesla crash, I still feel safe in my self-driving car
by Vivek Wadhwa on 07-24-2016 at 12:00 pm

At first, the thought of letting my car drive itself seemed rather frightening. But the highway was almost empty and the lanes were clearly marked, so I took the risk and engaged the autopilot function in my new Tesla Model X. Yet I couldn’t let go of the steering wheel. I didn’t want to put my life in the hands of software. This was two weekends ago as I drove to Big Sur, Calif.

The fear lasted about five minutes. Curiosity got the better of me and I let go of the steering wheel to see what would happen. The car continued to drive just fine; it didn’t need me. After a couple of minutes, the car beeped and displayed a message on the dashboard asking me to put my hands back on the wheel — a feature the automaker added to ensure drivers were in the front seat and attentive.

But 20 minutes later, I had one hand on the wheel and I was checking email with the other as the car did the driving for me. I did take full control when the road was narrow or the terrain was uneven, but by and large, I became as comfortable with the car’s autosteer function as I am with cruise control.

Yes, self-driving cars pose new risks, as evidenced by the recent fatal crash in Florida, when a Tesla in autopilot mode hit a large truck that crossed its path. The Tesla software cannot handle local roads, intersections or extreme hazards yet. There are limits to every technology. It is the same scenario as using cruise control on local roads — you just shouldn’t do it.

Three out of four U.S. drivers have the same fears I did, according to a AAA survey. The same survey revealed that only one in five would actually trust a driverless vehicle to drive itself with them inside. I have no doubt, however, that once they get behind the wheel of one, they too will be checking email as I did. They’ll feel as comfortable with software driving their cars as they are with software flying their airplanes.

Tesla calls its software “autopilot,” but it really is nothing more than cruise control on steroids. The autosteer function keeps the car in its lane, reads road signs, drives as much over the speed limit as you ask, and slows down or stops if there is a slower vehicle or obstruction ahead. If you want to overtake someone, you engage the turn signal, and the car will move itself to the adjacent lane when it can. I found this to be safer than changing lanes myself because of the blind spots. The advantage the Tesla has is that it can see in all directions at the same time. It literally has eyes in the back of its head.
I also learned how self-driving cars could prevent accidents when a car from the right jumped into my lane just as the setting sun blinded me. My car automatically slowed down and gave way. No, it didn’t honk.

Self-driving cars will improve our lifestyles and make the world smaller. They will prevent tens of thousands of fatalities every year. The best part is that they will do to pesky, dangerous human drivers what the horseless carriage did to the horse and buggy: banish them from the roads. Software malfunctions will surely cause unfortunate accidents along the way. There will also be ugly public debates, efforts by incumbent businesses to create legislative barriers, and a lot of confusion.
But the technology is coming — whether we are ready or not. And I for one can’t wait to receive the software upgrades that will let the car do all of the driving. I look forward to enjoying the scenery or working during my commute.

If political leaders and lawyers in the United States try to stop progress — as is very likely — other countries will still adopt the new technologies and take the lead. We will end up playing catch-up with the rest of the world and miss out on the most amazing transition of our lifetimes: into an era in which we become the drivers in driverless cars.

For more follow me on Twitter: @wadhwa and visit my website: www.wadhwa.com


Brexit impact on semiconductors

Brexit impact on semiconductors
by Bill Jewell on 07-24-2016 at 10:00 am

On July 1, Daniel Nenni posted his thoughts on the impact of Brexit (the vote by the UK to leave the European Union) on semiconductors. In general I agree with his points. Below is my take on the issue.

The long term impact of Brexit is uncertain. The UK will likely negotiate a trade agreement with the EU which includes free trade between the two entities but not the free movement of people. Scotland may vote to leave the UK in order to join the EU. A 2014 referendum in Scotland was close, with 45% voting to leave the UK. There is a possibility other EU member nations may vote to leave. Italy, France, Sweden, the Netherlands, Austria, Finland and Hungary are mentioned as potential candidates. However, opinion polls show a majority of people in each of these countries favors staying in the EU.

The International Monetary Fund (IMF) has lowered their outlook for World economic growth in 2016 and 2017 primarily due to Brexit. The July 2016 IMF forecast calls for World GDP growth of 3.1% in 2016 and 3.4% in 2017, each down 0.1 percentage points from the April 2016 IMF forecast. The most significant change is the UK forecast with 2016 GDP growth down 0.2 percentage points and 2017 down 0.9 percentage points compared to the April forecast. Other forecasters are more pessimistic. Scotiabank expects zero UK GDP growth in 2017. The table below compares recent forecasts for UK GDP growth with forecasts made prior to the Brexit vote.

[TABLE] align=”center” border=”1″
|-
| colspan=”10″ style=”width: 675px; height: 23px” | United Kingdom GDP Growth Forecasts
|-
| style=”width: 132px; height: 29px” | Source
| style=”width: 54px; height: 29px” | Date
| style=”width: 60px; height: 29px” | 2016
| style=”width: 54px; height: 29px” | 2017
| style=”width: 60px; height: 29px” | Date
| style=”width: 66px; height: 29px” | 2016
| style=”width: 66px; height: 29px” | 2017
| style=”width: 60px; height: 29px” | Change
| style=”width: 72px; height: 29px” | 2016
| style=”width: 51px; height: 29px” | 2017
|-
| style=”width: 132px; height: 29px” | IMF
| style=”width: 54px; height: 29px” | July
| style=”width: 60px; height: 29px” | 1.7
| style=”width: 54px; height: 29px” | 1.3
| style=”width: 60px; height: 29px” | April
| style=”width: 66px; height: 29px” | 1.9
| style=”width: 66px; height: 29px” | 2.2
| style=”width: 60px; height: 29px” |
| style=”width: 72px; height: 29px” | -0.2
| style=”width: 51px; height: 29px” | -0.9
|-
| style=”width: 132px; height: 29px” | Scotiabank
| style=”width: 54px; height: 29px” | July
| style=”width: 60px; height: 29px” | 1.3
| style=”width: 54px; height: 29px” | 0.0
| style=”width: 60px; height: 29px” | Feb.
| style=”width: 66px; height: 29px” | 2.0
| style=”width: 66px; height: 29px” | 1.9
| style=”width: 60px; height: 29px” |
| style=”width: 72px; height: 29px” | -0.7
| style=”width: 51px; height: 29px” | -1.9
|-
| style=”width: 132px; height: 29px” | Focus Economics
| style=”width: 54px; height: 29px” | June
| style=”width: 60px; height: 29px” | 1.4
| style=”width: 54px; height: 29px” | 0.3
| style=”width: 60px; height: 29px” | April
| style=”width: 66px; height: 29px” | 1.9
| style=”width: 66px; height: 29px” | 2.1
| style=”width: 60px; height: 29px” |
| style=”width: 72px; height: 29px” | -0.5
| style=”width: 51px; height: 29px” | -1.8
|-
| style=”width: 132px; height: 29px” | PWC
| style=”width: 54px; height: 29px” | July
| style=”width: 60px; height: 29px” | 1.6
| style=”width: 54px; height: 29px” | 0.6
| style=”width: 60px; height: 29px” | March
| style=”width: 66px; height: 29px” | 2.0
| style=”width: 66px; height: 29px” | 2.2
| style=”width: 60px; height: 29px” |
| style=”width: 72px; height: 29px” | -0.4
| style=”width: 51px; height: 29px” | -1.6
|-

The bigger question is whether Brexit is a sign of changing attitudes toward free trade. Trade has been a major issue in the United States presidential campaign. Republican nominee Donald Trump wants to renegotiate the North America Free Trade Agreement (NAFTA) among the U.S., Canada and Mexico. Trump is opposed to a U.S. trade agreement with Central America (CAFTA) and opposed to the Trans-Pacific Partnership (TPP) agreement among the U.S., Japan, Mexico, Canada, Chile, Peru, Australia, New Zealand, Singapore, Brunei, Vietnam and Malaysia. Trump wants aggressive trade negotiations with China, threatening punitive tariffs on U.S. imports from China. Expected Democratic nominee Hillary Clinton is also opposed to CAFTA and TPP, even though these are supported by President Barack Obama. Clinton has also questioned the effect of NAFTA, even though it was championed by her husband Bill Clinton when he was President.

How does all of this affect the semiconductor market? The direct effect of Brexit is not significant. The UK is not a meaningful producer or user of semiconductors. According to data collected by the United Nations (UN) UK imports of semiconductors were $2.4 billion in 2014 and exports were $4.7 billion, a small amount compared to the global market of $336 billion. An indirect effect is Japan’s SoftBank Group has agreed UK’s ARM Holdings PLC for US$32 billion. ARM designs and licenses processors which are in 95% of the world’s smartphones, according the Wall Street Journal. SoftBank said Brexit did not affect the ARM bid, but it did make it cheaper as the Japanese yen has appreciated against the British pound since the Brexit vote.

If Brexit is a sign of moves toward more restrictive global trade, the impact on the semiconductor market could be substantial. Global trade has been important to semiconductors for decades. Semiconductor companies were among the first U.S. companies to move some manufacturing overseas. One of the pioneer companies, Fairchild Semiconductor opened assembly sites in Hong Kong in 1961, South Korea in 1966 and Singapore in 1968. Fairchild was soon followed by Motorola opening a plant in South Korea and Texas Instruments opening plants in Taiwan and Singapore. Today wafer fab capacity is spread around the world. IC Insights shows capacity distribution at the end of 2015 as:

[LIST=1]

  • Taiwan 22%
  • South Korea 21%
  • Japan 17%
  • North America 14%
  • China 10%
  • Europe 6%
  • Rest of World 10%

    The importance of global trade to the semiconductor market is evident by imports and exports by country. UN data for 2014 semiconductor imports and exports by key countries are shown in the chart below. The data for Taiwan is from the World Trade Organization (WTO) since it is not a UN member. Import and export data can be misleading as some countries such as Hong Kong and Singapore are trading hubs, with most of the semiconductor imports later included as semiconductor exports.

    China is by far the largest importer of semiconductors at $241 billion in 2014 (blue bars on right). Hong Kong and Singapore are the next largest sources of imports, but most of these are later exported. The U.S., Taiwan, South Korean, European Union (EU) and Japan are all significant importers, ranging from $27 billion to $38 billion. The major semiconductor exporters (red bars on left) correspond with the major wafer fab locations – China, U.S., Taiwan, South Korea, Japan and Europe (excluding the trading hubs of Hong Kong and Singapore). The total of the countries shown is $570 billion in imports and $482 billion in exports. The world semiconductor market was $336 billion in 2014 according to WSTS. Thus many semiconductors pass through multiple countries on their journey from the country of manufacture to the country of final consumption.

    The U.S. Semiconductor Industry Association (SIA) supports the Trans-Pacific Partnership Agreement (TPP). The SIA states “International trade is vital to the U.S. semiconductor industry and the American economy as a whole.” The TPP would simplify trade among the member nations while including provisions to protect intellectual property.

    Trade agreements are controversial. Opponents in high wage countries fear the movement of jobs to lower wage countries. Proponents argue increased trade between nations increases overall economic activity and adds jobs. Global trade has been an important driver of the semiconductor industry. Hopefully Brexit is not a sign of future trade barriers which may hamper the growth of the industry.


  • Why did Softbank pay so much for ARM? Because it’s worth it

    Why did Softbank pay so much for ARM? Because it’s worth it
    by Gus Richard on 07-24-2016 at 7:00 am

    Softbank’s acquisition of ARM Holdings was not only unexpected, but also the valuation was astonishingly high. Softbank is acquiring ARM for $32.2B or 23x CY15 revenue and 46x CY16 earnings. This was a 46% premium to the prior day’s closing price before the announcement of the acquisition. The questions to ask are: Why is Softbank buying ARM, and why are they paying so much?

    Softbank is primarily a communications and media company. It is accustomed to making large investments with long-term pay offs that create an annuity. For example, investment in communication infrastructure and wireless spectrum are large up front investments that provide long-term dividends and cash flow. Softbank’s long-term goal is to become a technology company that focuses on information technology. The importance and growth of artificial intelligence and the automatic accumulation of knowledge frames the company’s long-term vision. Over time, ARM could have the potential of leveraging Softbank into a leadership role in the data economy along with the Magnificent Seven: Google, Facebook, Amazon, Microsoft, Alibaba, Baidu, and Tencent.

    In the PC era, technology was dominated by Microsoft and Intel who together captured a majority of the profits. The hardware and software in the current mobile technology era is dominated by ARM and Android with Apple, Google and Facebook who monetize this generation of technology. I would argue that the next era will be the data era which will be dominated by connecting things commonly known as the “Internet of Things” (IoT). IoT’s dominant microprocessor architecture and operating system are yet to be determined; however, increasingly the value creation will be in the data generated by the large number of connected things. In the coming years, the data economy is estimated to be worth $1.6T. This includes value from productivity improvements, proactive maintenance, and ad placement on the web. ARM is the entry price for Softbank’s participation in the data economy.

    In 2015, ARM had a 32% unit share in its served markets. According to the company, this is up from 22% in 2011. ARM base chips’ market share as measured by revenue in 2015 was roughly 50% surpassing Intel’s overall processor market share. Softbank has indicated that it would double ARM’s R&D budget, which would accelerate its roadmap and drive further market share gains, thus squeezing out all other architectures except Intel. This would result in unit market share of processors in the 60-80% range over the next 10 years and would be equivalent to 60-80B ARM based chips a year. I believe roughly 20-30B of these chips would have an IP address and would be connected to the Internet. With this level of dominance, it would take decades to displace ARM due to the massive amount of code written for ARM’s instruction set. This dominant position would potentially provide Softbank with tremendous leverage in the data economy. What is yet to be determined is how Softbank would monetize this market position.

    ARM is the processor of choice in IoT applications along with Synopsis’s ARC processor. ARM is also developing an operating system and platform for the IoT, called mbed. With control and/or over site of the architecture, operating system, and platform specifications of IoT, ARM would be positioned to extract more profit out of the proliferation of connected things. In simple terms, if Softbank could generate $1 of recurring revenue for each ARM processor connected to the Internet in 2025, this would generate roughly $30B in revenue per year far surpassing ARM’s current revenue of $1.5B.

    Here are a few possible avenues for monetizing ARM’s architecture:
    [LIST=1]

  • Embed blocks of circuitry that could be turned on remotely. One example would be a security functional block. Softbank could share revenue with chip and/or system providers generated from these features, which would be very profitable for Softbank, the chip vendor, and OEM.
  • ARM’s architecture by 2025 will dominate network infrastructure, mobile phones, consumer products, auto and IoT, potentially creating pathways to access vast amounts of data. Data is the grist for the artificial intelligence and data economy mill and will only increase in value.
  • Alibaba Group and SoftBank are forming a joint venture to launch a cloud computing service in Japan. This might also provide an avenue for monetizing ownership of the ARM architecture.

  • Loving it when a Qualcomm plan comes together

    Loving it when a Qualcomm plan comes together
    by Don Dingee on 07-22-2016 at 4:00 pm

    Corporate layoffs are always a touchy subject. I think that’s because there is skepticism that one round of layoffs can turn into two, then if business still doesn’t improve the spiral accelerates into more rounds. Too many rounds indicate management didn’t really have a clue what was going on in the business, instead trying to placate shareholders with action.

    “Rightsizing” done right involves more of a strategy, with personnel actions combined with other financial steps plus sales and marketing actions to restart top-line growth. (Tom Peters: “You can’t shrink your way to greatness.”) The trick is to take out enough people, once, and realign the entire organization and incentives around that new figure.

    About this time last year, instead of celebrating its 30th anniversary, Qualcomm dropped a 15% workforce reduction as part of its strategic realignment plan. It’s been hard to tell if that worked. New data in Qualcomm’s 3Q16 earnings release shows signs that the SRP initiatives are getting traction. A presentation released on July 20[SUP]th[/SUP] outlines progress on the SRP:

    1) On track for $700M cost savings in FY16, $100M more than the original estimate.
    2) Reaffirmation that the current corporate structure (with both licensing and product activity) delivers better value
    3) $5.9B returned to shareholders in the first 9 months of FY16 (dividends plus repurchases)
    4) Director turnover – 7 retirements, 3 additions – reduced average tenure to about 5 years
    5) Performance-based executive compensation steps implemented
    6) Focused investments, including completing the CSR acquisition and setting up a JV with TDK for RF front-end solutions

    It always helps to issue conservative quarterly guidance and then outperform it. Qualcomm hit the top end of their revenue guidance at $6.0B, and exceeded unit shipment guidance by 6M parts at 201M. Combined with a better cost story, EPS outperformed guidance by 16 cents.

    A few charts portray an interesting story. This is MSM chip shipments by calendar year:


    Next is 3G/4G device shipment estimates. The fine print on this slide is a chilling note given the claims of international “cheating” heard in Cleveland last night; here it is verbatim:

    “Global 3G/4G device shipments represent our estimate of CDMA-based, OFDMA-based and CDMA/OFDMA multimode subscriber devices shipped globally, excluding TD-SCDMA devices that do not implement LTE. We continue to believe that certain licensees in China are not fully complying with their contractual obligations to report their sales of licensed products to us, and certain companies, including unlicensed companies, are delaying execution of new license agreements. As a result, we do not believe that all global 3G/4G device shipments are currently being reported to us.”


    The last table affirms that while device shipments are trending up, device ASPs (the basis for Qualcomm royalty payments) are headed down. That could represent a mix of lower priced smartphones in Asia, cheaper IoT devices, and a maturing of older product lines.


    Qualcomm’s 4Q16 guidance has yearly revenue in the range of a 1% decrease to a 14% increase – and I’m guessing it will be closer to that top figure. EPS will come in anywhere from 15 to 26% improved. They resisted the calls to split the company and instead laid in a plan to fix the current business while still investing in the future; their R&D and SG&A expenses were $7.8B in FY15 and are expected to drop only 3 to 5% this year.

    I’m not trying to sell stock here. (Again, I don’t own shares of QCOM, or anything else I write about.) To me this is a refreshing story about a firm that took decisive, thoughtful, multi-faceted action and may have turned the corner. We could have a conversation on near-term versus long-term, but it’s much easier to execute a long-term strategy when the building is not on fire and people aren’t fearful for their jobs. I’m sensitive to the human costs that came with a 15% reduction, but I’m also mindful of the economic benefits a healthy Qualcomm can deliver – for themselves and the electronics industry at large.

    Do you have your copy of “Mobile Unleashed” yet? Chapter 9 is dedicated to the origin story of Qualcomm, and Chapter 10 delves into the Chinese contingent of ARM licensees.


    Reducing Data Centre Cooling by 40%

    Reducing Data Centre Cooling by 40%
    by Daniel Payne on 07-22-2016 at 12:00 pm

    Living in Oregon has many benefits, including access to cheap electricity thanks to the plentiful river systems that provide hydro power and a growing green power business fueled by wind and sun. Many of the world’s largest data centers are located in Oregon for access to this cheap electricity, and Google has a sizable investment in the Dalles, Oregon.

    I’ve learned that the racks of servers found in a data center generate a lot of heat, so that keeping all of that electronics cool does take up a lot of energy itself. The bright engineers at Google decided that one way to reduce cooling costs would be to analyze the data for cooling by using an AI-based system known as DeepMind. What that yielded was a surprising 40% reduction in the cooling costs.

    Another unique decision by Google is to use renewable energy for their data centers as another way to reduce emissions into the environment. The cooling in a data center uses big industrial equipment:

    • Pumps
    • Chillers
    • Cooling towers

    What’s so difficult about cooling a large data center? It turns out that the data center responds dynamically to requests by users, so the actual servers don’t have a static profile, rather the power and therefore cooling bounces around a lot. The interaction between servers, cooling and demand by users is sophisticated and non-linear, so trying to use a traditional engineering formula or your own common sense doesn’t really help to optimize the cooling challenge. Cooling systems also don’t respond instantly, there is a certain lag time to get started, reach a level, or ramp down. The physical plant at one data center may be quite different from another data center, so an approach must be taken that is based on each unique location.

    For the past couple of years Google has used a machine learning based approach to help operate data centers in a more optimal manner than before. The DeepMind system was used by researchers to improve the cooling efficiency by creating a system with neural networks that could understand the various operating scenarios and parameters that characterize a data center. An adaptive framework helped Google to learn the data center’s interactions.

    Past data was already available for analysis from all of the sensors inside a data center:

    • Temperatures
    • Power
    • Pump speeds
    • Setpoints

    This data was then used to train a collection of deep, neural networks. Optimization was focused on the Power Usage Effectiveness (PUE), which is the ratio of total building energy divided by the IT energy usage. Two additional neural networks were trained to predict the upcoming temperature and pressure over the next hour operating the data center. These predictions helped to simulate what actions were recommended from the PUE model, so that the cooling system operated within specifications.

    Here’s a plot showing the PUE value as a function of time, where a lower number is better because it saves power:

    When the Machine Learning (ML) starts we can see a very quick drop in PUE, which shows the 40% savings in energy used for cooling.

    Summary
    Google data centers are becoming even more energy efficient by using machine learning approaches and neural network modeling to reduce power consumption for cooling by 40%.

    Read the full blog about Google DeepMind and saving 40% on cooling costs here.