RVN! 26 Banner revised (800 x 100 px) (600 x 100 px)

IoT Application: Road Biking Fitness

IoT Application: Road Biking Fitness
by Daniel Payne on 08-10-2014 at 8:00 pm

Eleven months ago I started a fitness kick in order to lose some weight, get healthy and have more energy, so I picked a familiar activity, road cycling. Being an engineer I have always loved measuring things, like my speed and distance, however I had an old-fashioned cyclocomputer called the Cateye Velo 2. This device connected to the handlebars and a wire went down the front fork to a sensor that picked up the magnetic signal from a spoke-mounted magnet.

Being some 10 years old, there was no mechanism for the Cateye Velo 2 to transfer the cycling data to a computer or the Internet, so I went searching for something high-tech that would connect my ride data to the web. My first approach was to use an Android-based cell phone with GPS enabled and a free app called Strava.This worked pretty well to show me a map of the route, speed and elevation.

The Strava app worked fine on my Samsung Galaxy Note 2 phone, and I just had to be patient waiting for the GPS signal to lock before starting a new ride. When the ride finished I pushed one button in the app to sync my ride data by WiFi to the Strava web site.
Related: Non-volatile Memory in the Internet of Things

Somehow I wasn’t content with just velocity, because I wanted to know my cycling cadence and heart rate too. My next approach was to buy a wireless device and I looked at several companies:

Garmin was too expensive for my budget, Trek was possible, Sigma was a brand not familiar to me, so I ended up with the Cateye Stealth 50. This device uses GPS to capture my location, along with wireless sensors for: speed, cadence and heart rate.

Being curious, I took the device apart to see what was inside, which of course voided my warranty.


Related: ARM and the Internet of Things

To the right is the lithium ion battery, separated from the main PCB by a black plastic housing. The PCB on the left has four main chips on it, none of them had corporate markings. From the Cateye website I learned that one of those chips is a 4-bit micro-controller. I can guess that this device has chips for:

  • 4-bit micro-controller
  • LCD display driver
  • ANT+ protocol
  • USB connectivity

The four shiny posts on the bottom of the PCB are the connectors to dock the Stealth 50 to a USB cradle, used after a ride to load your data to a computer, then upload to Strava or Cateye’s web site for analysis. The retail price of this device is $150.00, so I’ll also guess that the margins are also quite high with such a small Bill of Materials. Since were talking about making measurements a few times per second, the chip technology could probably be a mature process node like 180nm, so you don’t really need bleeding edge technology to make a successful IoT device.
Related: The Internet of Things

All four companies listed above support the ANT+ protocol for ultra low-power wireless device monitoring, and there’s even an industry association to promote it for a diverse set of uses: runners, swimmers, cyclists, exercise equipment, hiking and medical.

For my bike I added a Garmin sensor to monitor both speed and cadence, it attaches near the rear wheel, and uses the ANT+ wireless standard.

While at the Bike Gallery in Beaverton I purchased an ANT+ compatible heart rate sensor from Bontrager (Trek brand).

Since Cateye is a Japanese company their user manuals take some interpretation, so it was much easier for me to visit the USA web site and watch a how-to video to get all my gear talking and working together.

The fun part is actually going for a ride and being able to view in real time my: Speed, Cadence and Heart Rate. After the ride I can see all of the numbers or graphs in Strava:

The above graphs show the elevation of my route, speed, estimated power, heart rate and cadence. For the time point at the cursor I was gliding down a hill at 40.7 mph, while not pedaling so the cadence was 0, and a heart rate of 156 bpm. On that ride I averaged 90 rpm for cadence, 153 bpm for heart rate, and 20.7 mph for speed. Now that’s a lot of data for a cyclist to consume. The only other technology gizmo that I could add would be either a power meter crankset or pedals, but at prices between $995 and $1,995 that’s way too rich for my budget.

Summary

The IoT as a cyclocomputer and wireless sensors has many vendors to help you measure your fitness level and achievements. Standards like ANT+ make a lot of sense for both consumers and vendors, instead of taking a proprietary approach and you really can mix-and-match between them like I did. On the software side the two main web-based providers are Strava and MapMyRide, I started with MapMyRide and then moved over to Strava where the more competitive road cyclists hang out. All of this is made possible at an affordable price because of our semiconductor ecosystem powering the electronics.

Hopefully I may have inspired some of you to get fit too. I’ve lost 30 pounds in 11 months, and am at 170 pounds today because of my cycling fitness efforts. I’ll never go back to my old, sedate lifestyle again.


My son and I just finished the Portland Bridge PedalRide

Semiconductor Revenue Trends

Semiconductor Revenue Trends
by Peter Gasperini on 08-10-2014 at 9:00 am


Image Source: Wikipedia

Nescire autem quid ante quam natus sis acciderit, id est semper esse puerum. (Not to know what happened before you were born is to remain forever a child.)Cicero

2014 is destined to be a pivotal year for Silicon Valley and High Tech in general. End user markets have been stagnating or declining over the last several years – most notably, the HDTV and consumer PC markets. Even high flying segments such as tablets and smartphones have been slowing markedly, with the CEO of Best Buy complaining about collapsing tablet sales and cellphone service providers urgently promoting ever more aggressive smartphone pricing and service bundles to reignite momentum.

Yet according to most analysts, this is supposed to be a year of energetic recovery, in the way that 2012 and 2013 weren’t. The SIA has been enthusiastically proclaiming that they expect a new record year of revenues in 2014 for the semiconductor industry.

The situation is indeed puzzling. After all, the SIA reported 2013 was a record revenue year for the chip industry worldwide, with 4.8% growth to $318B. Yet at the next layer of detail, the news is less reassuring. The logic portion of semiconductors showed only a 0.4% uptick, with the overwhelming majority of revenue gains going to DRAM and Flash – in particular because of the memory chip supply shortage that carried over from 2012. Reports for 2014 have so far been encouraging, but the industry has yet again experienced memory shortages for at least the first half of this year.

The gross numbers are clearly not sufficient to gain a thorough understanding of how the chip business is faring. A more accurate assessment is possible by segmenting the industry according to markets served and technologies on offer, with a distinct prejudice against firms in the memory sector. To that end, I’ve assembled a portfolio of ten companies that span the gamut of market segments served by silicon – the three C’s (communications, consumer and computing), mobile computing (tablets & smartphones), ISM (industrial, scientific and medical), automotive and, finally, mil/aero. Technologies included are a broad mix of SoC, programmable logic, microprocessors and even systems & software with several systems houses thrown into the mix. From this selection, one can get a snapshot of the relative health of the logic portion of the chip sector, major systems markets served and the entire High Technology value chain.

What’s also vital for this data to have relevance is a historical perspective. To serve that purpose, I’ve compiled financial reporting information all the way back to Q1 2008. Please note that corporate financial statements are very often non-GAAP to exclude one-time charges or certain unpleasant liabilities (as well as to provide maximum room to spin the message for investor audiences.) Regardless of that fact, the numbers are still extremely useful for illustrating long term trends, and are presented below, with $B on the Y axis.

The amount of information on this graph is a bit overwhelming, but there’s a couple of things that leap out from it:

[LIST=1]

  • A couple of these companies show regular Q4 holiday season spikes – Apple, Microsoft and (strangely enough) IBM stand out in this.
  • The semiconductor firms are, with the notable exception of Qualcomm, relatively flat.
  • The venerable systems houses – IBM, HP and Cisco – are either flat or, rather alarmingly, trending down quite steadily over the long term.
  • There most certainly has not been a broad based recovery with vigorous growth since the 2008 Wall Street financial crisis. This suggests more forces are at play.

    Some of these companies are vastly larger than others and would benefit more from a segmented charting and analysis, grouping companies together by type and/or markets served. I’m doing that currently and offering the results on http://vigilfuturi.blogspot.com. For those who are interested in getting access to a spreadsheet of the source data, please let me know in the comments and we’ll figure out together how I can get the data to you for your own analyses.


  • Analog Model Equivalence Checking Accelerates SoC Verification

    Analog Model Equivalence Checking Accelerates SoC Verification
    by Pawan Fangaria on 08-09-2014 at 7:30 pm

    In the race to reduce verification time for ever growing sizes of SoCs, various techniques are being adopted at different levels in the design chain, functional verification being of utmost priority. In an analog-digital mixed design, which is the case with most of the SoCs, the Spice simulation of analog components is the limiting factor because it is inherently slow and cannot handle large designs. A mixed-signal simulator which integrates Spice solver with event-based (for digital) solver can be used to improve the overall speed (better than Spice) while keeping the accuracy at Spice level; however the speed is still substantially less than a pure event-driven simulator and capacity to handle large analog content is still a limitation. So, what’s the next alternative?

    We do have one where analog behavioral models can be used in place of analog IPs or components and can be simulated along with digital with event-based solvers at the same speed as digital without any major capacity limitation. An analog behavioral model is the abstract representation of its analog IP or component; the behavior is captured using either RealNumber or logic data types in HDLs. Since the actual implementation of the analog IP or component is done separately through schematic and layout in the design process, it’s necessary to check the equivalence between the actual implementation and the behavioral model.

    In the framework of analog behavioral model validation, verification planning is very important which brings together all stakeholders from analog, digital and verification teams, sets objectives of equivalence validation and establishes a continuous communication channel. The equivalence between behavioral model and circuit implementation must be ensured by verifying it every time any change in specification occurs which drives change in circuit implementation. The verification plan must contain all important criteria such as list of features and their tests, testplan, coverage, equivalence validation methods, review procedure, reporting, exception handling and verification closure criterion.

    In the simulation environment, every test has to be simulated twice; first the Spice view of analog IP and then the behavioral model as DUV (Device Under Verification). The test scenarios are derived from the verification plan and EqVC (Equivalence Validation Component) is placed in the testbench environment to capture information that is used to compare if simulation of both types of DUVs produce similar results and decide pass/fail for each test or waive off certain failing test depending on the closure criteria mentioned in the verification plan. The comparison technique varies with the type of EqVC. The testbench can be setup either in analog-on-top (i.e. schematic driven environment) or in digital-on-top (i.e. HDL based environment) fashion.

    This methodology of equivalence validation has been used and presented jointly by Mentor Graphicsand ST Microelectronicsin DVCon 2014. They used digital-on-top testbench environment and mixed-signal simulator which could handle both behavioral model as well as Spice view of analog IP without any change in testbench. To reduce simulation time, focus was on testing features relevant for functional verification; further, model with multiple functions was broken down into smaller simpler models, which aligns with the philosophy of analog circuits being modular, consisting of sub-circuits.

    There are three different types of EqVCs

    For quick equivalence validation, ‘waveform compare’ method is used which can be quickly setup with a waveform tool like EZWave for comparing continuous waveforms and automating post-processing of results for reports, such as pass/fail, through scripting. This method is good for simple models such as BIAS and PLL models of HDMI IP.

    Above is a snapshot of waveform mismatch between analog and behavioral model. The differences shown in red can be studied and used to fine tune tolerances and waiver criteria.

    The Assertion Based Verification (ABV) is similar to the way it is used in digital circuits, that is by adding assertions in the design description to capture the design intent and verify that the intent is implemented correctly. QuestaADMS extends SVA (System Verilog Assertion) bind scope to analog objects as well. This methodology improves design quality and verification productivity by increasing observability. It’s well suited for digital-on-top verification; the testplan acts as an executable verification plan and is linked to functional checks (in the form of assertions) from simulation results. A library of commonly used checkers and monitors such as monotonicity of signal, signal crossing a threshold etc. can be reused with different models whereas specific assertions have to be written afresh based on the functional specs.

    Above is an example of a testplan for a voltage regulator. The QuestaADMS allows use of scripts to link testplan and merge UCDB. The assertions from behavioral model simulation as well as Spice DUV are merged separately and then compared to ascertain failing testcases in the behavioral model.

    The Advanced Verification method is based on UVM, a standard methodology promoted by Accellera, based on reusable class libraries. This method includes constrained random stimulus to increase verification coverage and supports coverage driven verification. UVM agents are extended for analog specifications to increase test scenarios and measure coverage. The basic test methodology remains identical to ABV.

    Read more on coverage driven verification through UVM

    ST has successfully used this validation environment for multiple of their IPs. A mix of EqVC types can be used; ‘waveform compare’ method for simple models at leaf level and ABV/UVM method for complex models and interconnected models at the top level. QuestaADMS supports RealNumber data types in Verilog-AMS, VHDL and SystemVerilog and supports the above verification methodologies for low power and mixed-signal simulations. More detailed information can be found in a whitepaper at Mentor website.

    More Articles by Pawan Fangaria…..


    Speeding up IP and Data Management

    Speeding up IP and Data Management
    by Daniel Payne on 08-09-2014 at 8:00 am

    IP and Data Management (DM) for SoC teams has gradually moved from ad-hoc approaches using simple Excel spreadsheets, to home-grown software that is specific to a project or company, and finally to commercially supported tools. One such commercial toolset for IP lifecycle management is from Methodics, named ProjectICwhere you can create, update and track workspaces using an open architecture, interfaces and data:

    It’s possible for a modern IC design team to be working with an 80GB workspace, which can take a long time to sync using DM tools like Subversion, Perforce, Git and others. To speed up this sync process there are a couple of approaches available:

    • Intelligent IP Caching – Project IC ensures that the data designers need is located as close to them as possible through an onsite IP cache, and that there is no unnecessary duplication of data. Only a subset of the data actually needed for editing is held in the user workspace, then visible for data management sync operations, providing a speed up for syncing and smaller file sizes. I blogged about this topic a bit back in January 2014.
    • Accelerating Underlying Storage – Since project IC is built on top of open interfaces and standards they are able to take advantage of any new technology that can speed workspace creation. An example of this speedup is a collaboration Methodics have with NetAppthat uses the FlexClone data provisioning technology, allowing 100GB workspaces to be created in less than one minute, versus having to wait hours when using a typical Subversion or Perforce operation.

    The FlexClone technology can create an exact, virtual copy of huge data volumes and files without buying additional storage space. This partnership between Methodics and NetApp looks very promising, and even Perforce is using and recommending NetApp to their customers doing hardware design, games development and other users of large data. Perforce joined the NetApp Alliance Partner program last year in October.

    Engineers at Methodics have taken the NetApp technology and added it directly inside of the ProjectIC software, speeding up IP and data management tasks. Other IP and DM vendors tend to build their own databases and version control systems, which kind of locks you into their tools, so that when they try to boost sync speed with NetApp it requires custom OS updates. The way that Methodics is using NetApp to dramatically cut down the workspace sync time is really transparent to the IC designers, yet does not lock your data into either companies tools.

    Further Reading

    If you’re interested in learning more about IP and DM, then consider these resources:


    Should we pay the price of Innovation?

    Should we pay the price of Innovation?
    by Eric Esteve on 08-08-2014 at 8:00 pm

    I agree that this question sounds stupid: nobody is forcing me to buy an innovative product, or even a gadget, if I don’t want to pay a high price, I just don’t buy the product. But it seems that some people don’t really think that way. The story is related to Qualcomm sales in China, and recently announced partnership with SMIC…

    The Partnership (the fact)

    From the joint Press Release: SAN DIEGO – July 03, 2014 – Semiconductor Manufacturing International Corporation (“SMIC”; NYSE: SMI; SEHK: 981) and Qualcomm Incorporated (NASDAQ: QCOM), have announced that SMIC and Qualcomm Technologies, Inc., a subsidiary of Qualcomm Incorporated, are working together in connection with 28nm process technology and wafer manufacturing services in China to manufacture Qualcomm® Snapdragon™ processors. Qualcomm Technologies’ Snapdragon processors are purpose built for mobile devices. SMIC is one of China’s largest and most advanced semiconductor foundries, and Qualcomm Technologies is one of the world’s largest fabless semiconductor vendors and a world leader in 3G, 4G and next-generation wireless technologies. This collaboration will help accelerate SMIC’s 28nm process maturity and capacity, and will also make SMIC one of the first semiconductor foundries in China to offer production locally for some of Qualcomm Technologies’ latest Snapdragon processors on 28nm node, both PolySiON (PS) and high-K dielectrics metal gate (HKMG).

    This PR sounds like both companies are enjoying a new partnership, maybe showing that one of the partners is getting higher benefit: “This collaboration will help accelerate SMIC’s 28nm process maturity and capacity, and will also make SMIC one of the first semiconductor foundries in China to offer production locally for some of Qualcomm Technologies’ latest Snapdragon processors on 28nm node…”. If you further analyze, “Qualcomm will help SMIC accelerate 28nm process maturity” sounds like the customer is devoting resources to help the supplier filling the technology gap with foundry competitors. If you prefer, this PR sounds like Qualcomm is paying an entry ticket to stay active and continue to sale Snapdragon on the Chinese market. Maybe this deal does not look any more like a win-win deal? The good question is to know why Qualcomm had to sign such a partnership?

    I found a possible answer in this article from Junko Yoshida, Chief International Correspondent, EETimesChina’s SMIC-Qualcomm 28-nm Deal: Why Now? “, here is an extract:

    Antitrust investigation in China
    Since China launched an antitrust probe into Qualcomm late last year, speculation abounds that Chinese authorities are probing ways to coerce Qualcomm into collaborating with their electronics industry.
    Qualcomm reportedly faces penalties that may exceed $1 billion. The National Development and Reform Commission (NDRC), China’s main planning body, raided Qualcomm’s Beijing and Shanghai offices last year.
    The NDRC has used the anti-monopoly law to target technology companies for practices that could lead to what it calls “unreasonably” high prices. In February, the Chinese regulator said it suspects Qualcomm of overcharging and abusing its market position.

    So the Chinese regulator (NDRC) considers that technology companies like Qualcomm are selling at “unreasonably” high prices. Let’s make a point: Qualcomm has invented and patented innovative modem techniques (CDMA and the like) for wireless communication, and these techniques have been selected by the telecommunication regulators in the USA (and other regions) to be at the hearth of the new standards. Qualcomm has a de facto monopoly, this is due to the international patent policy: every chip maker developing a modem has to pay a license and royalties to QCOM, and this gives a competitive advantage to Qualcomm when the company also develop modem IC. Qualcomm has been smart enough to also dominate the Application Processor market. The chip maker has just do a better job that TI, Nvidia, Marvell, Freescale… you name it. The equation is rather simple:

    Innovation (Patent) + Investment (IC design) + Roadmap = Strong Leader position

    As far as I am concerned, I don’t see any malfeasance in this strategy. We have seen in the past a high tech PC chip maker basing the company development, not only on a quasi-monopoly (leaving just enough room for a single competitor to survive, so the monopoly was not 100%), but also on anti-competitive practices (like paying back customers to make sure these will stay). Such a behavior has been sanctioned by the American law, and this was good decision. But the picture is completely different with Qualcomm. If you agree with the international patent policy, you must admit that a company cleaver enough to create innovation and turn it into a new technology and the related (IC) products should be in a position to harvest and get benefit from this innovation…

    Let’s make it clear: I have no negative a-priori against China. But I may have a certain reluctance when I see politician (from any country) trying to squeeze innovation. At the end of the day, SMIC will get benefit from this partnership, detrimental to TSMC, Samsung or GloFo, and detrimental also to innovation.

    Eric Esteve

    More Articles by Eric Esteve…..


    IBM thinks neural nets in chip with 4K cores

    IBM thinks neural nets in chip with 4K cores
    by Don Dingee on 08-08-2014 at 2:00 pm

    Neural networks have been the darlings of researchers since the 1940s, but have eluded practical hardware implementations on all but a small scale, or an enormous one given how many processing elements and interconnects are needed. To make significant brain-like decisions, one needs at least several thousand fairly capable cores and a massively configurable interconnect.

    An example of how much fun this problem is surfaced a couple years ago, when Google created a software neural network with 16,000 nodes and turned it loose on 10 million YouTube videos looking for cats. (No, they weren’t looking for the text “I can haz cheezburger?” That would be cheating. They were training on pictures of randomly selected cat faces.)

    One of the closest attempts at neural net hardware I’ve seen so far is based on Parallella, from Adepteva. The primary processor is a Xilinx Zynq with its FPGA fabric and a dual core ARM Cortex-A9, but there is also a 16 or 64 core Epiphany accelerator on board. Those cores are small, homegrown RISC designs running at 800 MHz with a mesh interconnect.

    It’s not hard to envision a cluster of maybe 64 Parallellas working together on a neural net problem – but there still needs to be a software paradigm to program that many cores effectively. OpenCL provides a solid, scalable environment for that task, able to deal with distributed processors in a heterogenous network.

    Reading through Nick Oppen’s blog gives a flavor of both the potential and the level of difficulty in such an exercise. Given enough time, programming acumen, and a few thousand bucks (Parallella boards start at $149) and space to set up 64 boards, it’s possible to get that working.

    If you’re IBM, and have a bunch of really smart researchers, and cost and development schedule is not an issue, you try to put that on one chip (minus the ARM cores and a fully programmable FPGA – think simple cores and configurable fabric). The project, working under the auspices of the DARPA SyNAPSE initiative, involves both hardware and software.

    The specs of the IBM TrueNorth chip just introduced are impressive: 1 million programmable neurons, 256 million programmable synapses, and 4096 neurosynaptic cores. The fascinating feature: there is no clock. The cores aren’t processors in the usual sense – they are event handlers, waiting dormant until a “spike” fires. Fabbed in Samsung 28nm, TrueNorth is IBM’s largest chip to date with 5.4B transistors, yet consumes less than 100mW.

    As researcher Dharmendra Modha shares, this isn’t an engine to run compiled C/C++. IBM matched the hardware to the Compass software simulator, a cognitive computing approach introducing the concept of corelets. This is sort of the network-on-chip on steroids; it abstracts and encapsulates intra-network connectivity and intra-core physiology, exposing external inputs and outputs which users connect to.

    IBM already has a 16-chip board running, and is targeting an unbelievable scale of 4096 chips in a rack – 4 billion neurons and 1 trillion synapses, in less than 4kW. They are also playing with a retinal camera from iniLabs which produces spikes instead of the traditional 2D imagery requiring DSP handling.

    We in EDA are somewhat slaves to the C/C++ or Java programming paradigm, because there is so much software and hardware IP and experience out there. As the Internet of Things unfolds, there will be new programming methods – much as FORTRAN and Pascal waned for general purpose use, I think C/C++ will eventually be supplanted as the underlying architecture morphs.

    Before that happens, developments like TrueNorth have to become a lot more cost effective, for sure; this is still research on a pretty intensive scale. However, did we imagine something as inexpensive as Parallella just a few years ago?


    Layout-aware Diagnosis

    Layout-aware Diagnosis
    by Paul McLellan on 08-08-2014 at 8:01 am

    Traditional test methodologies have been based on the functional model, that is to say the netlist. The most well-known is probably the stuck-at model which grades a sequence of test vectors by whether they would have managed to notice the difference between a fully functional design and one where one of the signals was permanently stuck at 0 (or 1). In some ways it is a crude measure, since many faults (such as two signals shorting together) don’t manifest themselves in precisely that way. But it turns out to be a lot better than might be expected. In the same way as code-coverage of Verilog (or C++) doesn’t guarantee correctness, for sure if a line of code is not executed you can’t tell if it is correct. In the same way, if a test sequence can’t detect a stuck-at fault then there are some malfunctions that wil make the chip fail but which the test sequence cannot detect.

    To take fault detection to the next level requires a better fault model but also requires taking into account more of the design than is available in just the functional netlist. For example, to model whether two signals are shorted together and whether the test sequence notices (produces different results if the short is modeled or not modeled) requires looking at which signals can potentially short. With the functional netlist there is no information and an explosively exponentially large number of signal pairs. But by taking the layout into account this can be pruned back to signals that are actually physically adjacent. Signal pairs that are never next to each other on the chip cannot short and so don’t need to be considered.

    Another problem that isn’t modeled by stuck-at is a signal that is open. If the fanout is more than 1 then there is the possibility of an open that leaves some of the fanout signals connected correctly, and others that are not driven at all (and will probably have a partial route from vdd to vss that is sometimes called crowbar current).

    Rating a test program for its effectiveness at finding faults is only one part of what test is about. That gives you the tools to improve the test program itself so that it works better.

    When there is a pattern of test failures then yield can potentially be improved by locating what is actually wrong and then fixing it. For example, if a layout hot-spot in optical proximity correction (OPC) is causing many failures then a minor change to how the reticle enhancement technology decoration is done (RET) may increase yield.

    Tessent has a tight connection between the layout engine and the logic engine, meaning that Tessent Diagnosis can remove more than 85% of all bridge suspects leaving just a few that are both logically and physically feasible. It can also reduce the bounding box of many signals that might have an open by noticing which signals are responding correctly and which are not, allowing the designer to home in on possible problem areas even though the net might be one that goes all over the chip.


    More articles by Paul McLellan…


    Who will Manufacture Apple’s Next SoC?

    Who will Manufacture Apple’s Next SoC?
    by Daniel Nenni on 08-07-2014 at 8:00 pm

    Just to review: The brain inside the current Apple iPhone 5s is the A7 SoC manufactured by Samsung using a 28nm process. The A6 (iPhone 5) and A5 (iPhone 4s) are based on Samsung 32nm. The rest of the Apple SoCs also used Samsung processes. I think we can all now agree that the coming Apple A8 SoC (iPhone 6) will use the TSMC 20nm process. In order to properly postulate which process the Apple A9 will use let me share with you my observations, opinions, and experience on the topic.

    Also Read: Will Intel Have a Bigger FinFET Market Share Than TSMC in 2015?

    In the beginning Apple started with Samsung as an ASIC customer where Apple did the preliminary design specifications and Samsung did the rest and delivered a completed chip. Over the course of the last ten years Apple evolved into one of the largest and most capable fabless semiconductor companies and now does everything required to get an SoC design into a foundry and the resulting chip into their products. In fact, Apple is now an “early access foundry customer” which means they are actively involved in early stage process development.

    The important question is: Why did Apple leave Samsung for TSMC?

    Apple is unique in that they release new mobile products in the fall of each year while competitors like Samsung release multiple products throughout the year. This ties the release of new foundry silicon to Apple’s new product releases since the volumes of wafers required are in the hundreds of thousands. Samsung’s delay from 32nm to 28nm was a big wake-up call for Apple. The iPhone 5 was supposed to contain 28nm silicon but clearly that did not happen which put Apple at a competitive disadvantage.

    Since TSMC is the only foundry to release a new process node in 2014 (20nm) with the wafer capacity to satisfy Apple (Apple has asked its suppliers to build between 70 to 80 million iPhone 6 handsets by the end of the year), Apple moved to TSMC for the A8. Moving to TSMC also clearly demonstrates that Apple is truly an independent fabless semiconductor company and can choose any foundry moving forward. This will enable Apple to play Intel, Samsung, and TSMC against each other for better wafer pricing, absolutely.

    Also read: What is the Latest in Mobile?

    Apple’s first FinFET SoC is a very difficult situation. I know for a fact that Apple carefully considered Intel 14nm, Samsung 14nnm, and TSMC 16nm. The key criteria here is the iPhone 6s Fall of 2015 ship date. Which means the design must be taped-out by the end of Q3 2014 for production start in Q2 2015. Based on what I know today here are scenarios I would like to offer up for discussion:

    Apple will NOT use Intel 14nm in 2015.
    Intel is still learning how to be a foundry and Apple is very demanding so there is a high element of relationship risk here. Apple is also VERY closely tied to ARM and Intel does not work with ARM on process development like TSMC and Samsung do. Intel 14nm also experienced big delays which increased the risk of missing the Apple Q2 2105 production start date.

    Apple will NOT use TSMC 16nm in 2015.
    TSMC 16FF was on track to be in production 1H 2015 but the process was further optimized to be more competitive with Intel and Samsung. The new TSMC 16FF+ process will not be in production until 2H 2015 which will miss the Apple iPhone 6s launch.

    Apple will NOT use Samsung 14nm in 2015. From what I understand today Samsung 14nm is still having silicon correlation problems. And as we have seen with Intel, yielding at 14nm is no small feat. The risk of missing critical wafer delivery dates here is very high.

    Apple WILL use TSMC 20nm in 2015.
    It is my understanding that the Apple A8 will have a dual core CPU running at a maximum of 2GHZ and will not have an integrated modem. Thus the room for an improved A9 20nm SoC is pretty big, especially if Apple is concerned about 14nm FinFET production delays.

    Bottom line:
    14nm FinFET technology is still evolving, 20nm technology has room for improved power consumption and performance, and 10nm is years away. For Apple the low risk scenario is: 20nm SoCs in 2014 – 2015, 16nm SoCs in 2016 – 2017, and 10nm SoCs in 2018-2019. Sound reasonable?

    More Articles by Daniel Nenni…..


    Hybrid Memory Cube and the Intel Knights Landing

    Hybrid Memory Cube and the Intel Knights Landing
    by Arie Lashansky on 08-07-2014 at 8:00 am

    While looking for information on a Xilinx Spartan 6 Project with DDR memory I came across a new type of DRAM called the Hybrid Memory Cube (HMC). The technology made me want to take a closer look:

    The Hybrid Memory Cube is like a stack of DDR chips stacked die on die using through silicon vias to interconnect the dies the bottom die in not a Dram but a logic Die.The major difference is that the block is not addressed in a way of Address and Data lines like the traditional Jedec memory but in a High speed serial IO like PCI Express.

    Think of the cube as Four High Speed Serial Links or 16 lanes. The Hybrid Memory Cube has far better bandwidith as it can be addressed by 4 different High speed links. Each serial lane can run at 10GBps, 4 lanes make a link of 40GBps, in a chip with 4 links it is 160GBps.

    From what I can see the (HMC) is is a game changer for memory:

    [LIST=1]

  • The footprint vs memory density is far smaller as 8 dies Stacked
  • The Logic Die at the bottom of the stack takes the memory loading away from the CPU
  • The PCB may be easier to Route as no need for all data address and clocks to have the same timing as the data is serial (PCI compaired to PCI Express).
  • The link between the memory and the CPU becomes more abstract as the CPU does not have direct control over the memory via a logic device at the bottom of the stack. I’m not sure this is an advantage and does not have a way to make one memory call a high priority on a different one. (Not a real world problem at such High speeds). The logic controller (Bottom die) can also change the order in which the memory responds to calls.
  • Concurrency as each lane can pass requests to the cube so at the same time the cube may be reading two different memory locations

    An example of where HMC will be used is The Knights Landing chip from Intel. Looking at Knights Landing lets say each Atom core is connected a lane that means 16 cores work on one shared address space.That I think may be the 4 memory chips in front .The two at the back I see may have each core connected to a link (4 lanes)
    http://en.wikipedia.org/wiki/Xeon_Phi 72 atom chips on one die at 14nm.
    Note 32 GDDRS both on front and Back of the Knights Corner below Knights Landing will not look like this with 6 HMC mounted on the interconnectsubstrate modules one chip does the work of the whole Board.

    Intel’s current chip is the Family Knights Corner, the same Basic Idea as Knights Landing but far less Features with DDR5 and not HMC.

    http://ark.intel.com/products/codename/57721/Knights-Corner


    https://www-ssl.intel.com/content/www/us/en/processors/xeon/xeon-phi-coprocessor-datasheet.html

    The only way I can see to use the HMC memory seems to be by using a FPGA solution. Both Xilinx and Altera have IP partners (see the Video’s Below):

    Micron Talk at Hot Chips: http://www.hotchips.org/wp-content/uploads/hc_archives/hc23/HC23.18.3-memory-FPGA/HC23.18.320-HybridCube-Pawlowski-Micron.pdf

    https://www.youtube.com/watch?v=QOYE56OCE3o
    (After minute 26:50 to 1:05 J. Thomas Pawlowski explains HMC)

    Xilinx Demo: https://www.youtube.com/watch?v=GYpJqDYpuG4

    Altera Demo: https://www.youtube.com/watch?v=oMsWRr0eBg4


  • What is the Latest in Mobile?

    What is the Latest in Mobile?
    by Paul McLellan on 08-06-2014 at 8:01 am

    Most of the results are in for mobile for last quarter, plus the earnings calls are all over. The picture is not pretty. The big picture is that low-cost Android-based suppliers, primarily in Asia, are starting to eat a lot of market share from Samsung (#1) and Apple (#2). There were about 295M smartphones shipped in Q2, a measly 2% up from Q1.

    Samsung just reported their worst quarterly profit for a couple of years and gave a guarded outlook for the rest of the year. Samsung probably has the most to lose from cheaper suppliers since they are supplying Android phones and have to be cost competitive with other almost identical phones being made by cheaper suppliers. In the quarter they shipped 74.1M units for about 25% market share. This is down from last quarter in unit terms since they shipped nearly 90M in Q1 (a huge drop) and, given their poor financial results, probably down a lot more in dollar terms due to margin pressure. They also stated that the outlook for the second half of 2014 is “challenging”.

    Apple shipped 35.2M units for a market share of 11% down from over 13% last quarter although up slightly in units. It is hard to know if this is problematic or not since Apple’s once a year product release schedule means that it is always weak at this time of year when they have to get rid of all the old inventory and buyers are all holding out for the shiny new model. But Apple has been losing market share (despite slight increases in units shipped) for some time now. But they remain differentiated from Android and the Apps/iTunes is a powerful disincentive to jump ship. Apple remains highly profitable, of course, taking a large part of the entire industry’s profits. With Samsung presumably losing money Apple could be making more profit that the entire industry, just as Samsung and Apple together used to.

    Talking of which, Apple has announced an iPhone related press event on September 9th. However there are plenty of rumors all over the web that the release of iPhone6 will not actually happen until October, but I have no idea if any of these are truly reliable. Also, Apple are supposedly pushing for a $100 price increase for iPhone6 that the carriers don’t like, at least in the US where phones are typically subsidized. I guess all will be revealed in a month.

    Next are the Asian premier league: Huawei, LG and Lenovo (without Motorola, see below). Different analyst houses have these in different orders, they are pretty close each with around 5% market share, 14-15M units.

    Then the Asian first division with Xiaomi, Coolpad and ZTE, also too close to call with around 4% each, around 11M units each.

    Xiaomi is 11M units for 4% market share (they claim over 5% and 5th place but that doesn’t seem to match anyone else, although they know their numbers better than the analysts). Not bad for a company that didn’t exist 3 years ago. Their sales are entirely in China, I believe, but they now plan to broaden out into Europe and other parts of Asia. Their trajectory is up and to the right and clearly they have a strong chance to end up several places higher.

    Sony is 9th, the lone Japanese entrant with about 8M units and 3% market share.

    Motorola had a great Q2 and crept into the top 10 at 10th place, which means that Microsoft/Nokia slips out to 11th. And just to remind you, Google sold Motorola to Lenovo but this result is just for Motorola since the deal has not finally closed yet due to regulatory review. When the deal closes Lenovo/Motorola should be #3 behind Samsung and Apple.

    I wouldn’t be surprised if the numbers in a few quarters time are:
    [LIST=1]

  • Samsung
  • Apple
  • Lenovo/Moto
  • Xiaomi
  • Huawei


    More articles by Paul McLellan…