webinar IPXACT banner

Have We Hit the Power Floor?

Have We Hit the Power Floor?
by Brian Fuller on 04-18-2015 at 7:00 am

As we celebrate the 50[SUP]th[/SUP] anniversary of the publication of Moore’s Law in Electronics Magazine (April 19, 1965), the industry finds itself in an increasingly costly global effort to keep transistor scaling on track. “Is Moore’s Law dead?” is a common question these days.

But practically speaking the doubling of transistor density every 18 months or so has been pushed off the front pages in the past decade by the crucial need to manage power.

Joel Hruska, writing in Extremetech, notes:

“One of the most striking characteristics of current semiconductor research is how completely the search for lower-power devices has subsumed the old clock speed obsession. 0W has become the new 1GHz. Performance, the old God of Computing is now merely an efficient means to achieve the lowest possible minimal power usage.”

In recent years, our industry has made enormous strides in power management, breathing longer battery life into smaller and smaller devices. But we’re starting to hear whispers of a question that echoes the one surrounding Moore’s Law: Are we approaching a power floor?

That’s one reason events like the annual Electronic Design Process Symposium (EDPS)—being held April 23-24—have become so critical to nurturing a thoughtful conversation about the future of low-power design.

“EDPS is a excellent opportunity for engineers to glean insight from industry experts on the latest techniques to design to increasingly stringent power budgets, consider emerging materials, and develop design flows that optimize implementation and verification for low-power systems,” said Arpana Dey, technical marketing director for standards at Cadence who serves as the 2015 EDPS chair.

The two-day event is being held at the Monterey Beach Resort in Monterey, Calif., and it will be day two that will offer a day-long deep dive into the most pressing engineering concerns surrounding low-power design.

Jim Kardach, director of integrated products at high frequency power conversion startup FinSix, will keynote first thing Friday morning on “Low Power Design, Standards and Evolution.” Kardach’s keynote will be followed by a series of presentations on “Low Power Technologies and Ecosystems,” a session chaired by Naresh Sehgal, senior program manager for Intel’s Imaging and Camera Group.

The Friday afternoon keynote features University of California San Diego computer science and engineering professor Andrew Kahng, speaking on “EDA/ESL Low Power Design Trends, ISTR/CAD and Tools.”

The low-power focus wraps up Friday afternoon with a panel session that I’ll moderate featuring Kahng, Kardach, Bernard Murphy from Atrenta, Steve Carlson from Cadence, Parasad Subramaniam from eSilicon , and Pat Sheridan from Synopsys. We’ll explore how much lower we can push power or whether we’ve hit the practical floor, especially for IoT designs.

Day one features a keynote from Tom Dillinger, Oracle CAD technology manager, which examines two of today’s key materials choices: fully-depleted silicon-on-insulator (FD-SOI) versus FinFETs. After Dillinger sets the stage, he’ll moderate a panel on the topic featuring Kelvin Low, senior director of foundry marketing with Samsung; Boris Murmann, associate professor of electrical engineering at Stanford; Marco Brambilla, director of engineering with Synapse Design; and Jamei Schaeffer, product line manager with GlobalFoundries.

The afternoon lineup on day one includes a variety of presentations in two sessions, one focusing on multi-die challenges and applications (chaired by EDA 2 ASIC President Herb Reiter), the second on hybrid virtual platforms (chaired by well-known industry analyst Gary Smith).

Here’s a link to the complete EDPS program and a link to the registration page. I look forward to seeing you there!


Successful Venture of an Indian Global VIP Company

Successful Venture of an Indian Global VIP Company
by Pawan Fangaria on 04-17-2015 at 10:00 am

It’s rare that we find a truly Indian-based company operating globally in the semiconductor space. Although the ‘gold rush’ towards IP development in the last decade initiated many IP start-ups in India, today we rarely find Indian IP company names which are shining in the global arena. The story of services companies is different, but a true product company is rare in India. A great product company needs real fundamental expertise to develop products and an inspirational leadership to strategise and execute in all aspects that can benefit customers across the world.

Although I already had a very good impression of SmartDV Technologies, headquartered in Bangalore, my actual insight about the breadth and depth of this company came into light when I read their last press releaseabout their new VIPs and subsequently had a few interactions with Harish Poojary, VP of worldwide sales and business development at SmartDV.

SmartDV was started in Bangalore, India in 2008 by a few highly talented and experienced individuals in complex ASIC design and verification domain. That was the right time when IP integration into large complex SoCs had picked up momentum. Deepak Kumar Tala, the Founder & CEO of the company had a great vision to grab this opportunity to automate verification through the use of VIP (Verification IP) and save enormous time spent in verifying complex SoCs. SmartDV management team feels proud of remaining ahead of the game by delivering a VIP for a protocol as soon as its final specification is released or even before that; a great strategy to help customers hitting the bull’s-eye during a smallwindow of opportunity in the SoC market.

SmartDV has a large portfolio of over 75 high-quality standard and custom protocol VIP products that work well with coverage driven verification flow. Each VIP comes with complete compliance test suite, comprehensive functional coverage model, and a rich set of customization and protocol checks to speed up the verification of designs. SmartDV VIPs are two to four times faster to compile and simulate compared to those offered by the competition. The test cases, coverage models and sequence libraries delivered with the VIP can be easily modified according to the customer environment to speed up the verification of SOCs at the customer end. The VIP portfolio is available in UVM, SystemVerilog, VMM, OVM, VERA, Verilog, Specman e, SystemC, or any other non-standard native environment of customer choice. No wrappers are added, thus enhancing the performance and simplifying debugging. The portfolio includes MIPI, Networking and SoC, Automotive and Serial Bus, and Storage VIPs, Memory models and Design IPs. Look at their product portfolio at their website here.

Above is an example of their recently released Networking and SoC VIP. The AMBA5 CHI VIP provides a smart and easy way to verify the ARM AMBA5 CHI component of an SoC. The AMBA5 CHI VIP is fully compliant with standard AMBA5 CHI specification and is supported natively in SV, UVM, OVM, VMM, Verilog, VERA, SystemC, Specman E, and any other language of customer choice. Similarly there are other VIPs, Design IPs, and memory models such as DDR4 and LPDDR4 that are conceived to be better than those available in the market.

What makes SmartDV so successful? Well, it’s the overall strategy, execution and operational excellence. However a couple of key points that emanate and significantly contribute in the success of SmartDV are – i) SmartDV’s own language and compiler technology that automated VIP development thereby enabling faster VIP development, ii) an automated development flow from specification to final product including documentation, and iii) a highly talented and experienced team that produces high quality products. The net result is “high quality products in short time with a smaller team size” said Deepak Kumar Tala. In fact, there were occasions when use of SmartDV VIPs exposed bugs in already taped-out chips; those bugs had passed the tape-out with VIPs supplied by other vendors.

As the engineering team continued developing their portfolio of IP products, business grew over the years without any marketing/sales efforts. In January 2014, SmartDV opened its office in San Diego, CA (USA)under the leadership of Harish Poojary. The strategy was to reach out to customers with better support, develop long term partnership, and profit from broader market opportunities. Today SmartDV has 80+ customers worldwide. They have plans to further expand in Japan and Taiwan. They are also working on automating memory models and simulation acceleration of VIP’s development.

SmartDV’s customers include top semiconductor companies in wireless, automotive, storage, memory and networking domains. Many of them have taped-out chips using SmartDV VIPs and are very impressed with the results.

SmartDV is exhibiting in DAC 2015. Visit their booth #514 to see the latest and greatest from this company.


Coventor, Lego and IoT in Denmark

Coventor, Lego and IoT in Denmark
by Paul McLellan on 04-17-2015 at 7:00 am

Coventor were in Copenhagen Denmark a few weeks ago at the Smart Systems Integration Conference to talk about MEMS and IoT entitled (take a deep breath) Towards a Lego Block Principle for Heterogonous Systems Design Including MEMS and Electronics—Choose and Put Together Fit. Since this seems to have become IoT week for me, without any deliberate plan, this seems very appropriate. After all, “Heterogonous Systems Design Including MEMS and Electronics” screams IoT. And also appropriate is to have the discussion in Copenhagen since Denmark is the home of Lego (in Billund, which is nowhwere near though). Also, Lego have to be one of the few companies that manufacture parts in semiconductor type volumes with a over a half-trillion bricks produced since the company’s founding in 1949. For some reason, Americans pluralize Lego to Legos even though the rest of the world appears not to (the other way around from mathematics which, in England at least, we abbreviate to maths while the US goes with math).

I was talking to someone yesterday, not at Coventor, who was telling me they had a lot of design technology for MEMS very early on and yet, even today, are not really in the market. I pointed out that history is littered with companies who entered some aspect of the MEMS market way too early and flamed out having run out of money. EDA is like that, more companies fail by being too early than too late.

Anyway, MEMS is clearly here now. It is now a multi-billion dollar industry and most of us have, in our pockets, accelerometers, gyroscopes and more inside our phones. Our cars know when they are crashing, our computers can detect they have been dropped, and many of us have a Fitbit band or something equivalent on our wrists (or sitting unused in a drawer at home). So what the somewhat confusing title, the discussion was about why MEMS is different. Why can’t we abstract design and do it in a manner similar to CMOS digital design? What is the equivalent of a transistor in MEMS? And if that approach is not going to work (after all, most digital designers rarely even see a transistor) can’t we at least have a Lego approach similar to how we assemble large SoCs out of blocks of pre-designed IP?

Of course, these days even analog CMOS designers don’t actually care that much about transistors, especially in the FinFET era. EDA tools with PDKs shield designers from such “details” and allow them to focus on functional blocks such as differential pairs, current mirrors, common-source, common-gate and common-drain circuits. So how hard can it be to have a standard MEMS flow?

This is not a new topic. Indeed, as Coventor’s panelist Gunar Lorenz pointed out on his blog entry, this was a topic he had discussed 15 years earlier with his PhD supervisor (and co-panelist) Richard Neul. So has Coventor solved the problem and convinced the MEMS design community that a library-based MEMS design flow is possible or even desirable? Not entirely. As a matter of fact there are MEMS designers who cherish and defend their freedom to draw about any idea which comes to mind in a layout editor without concern for restrictions imposed by a MEMS component library. To those of us with an EDA background this sounds just like the arguments we have had for decades: why can’t analog designers get with the program and automate their work…only even worse.

The introduction to the panel was given by Tobias Maier from Robert Bosch who emphasized the widely felt frustration about the lack of a standard design methodology for MEMS. The essence of the current MEMS design flow (or lack thereof) was nicely captured in a slide by Jörg Doblaski of XFAB, presented earlier in the conference:
Gunar isn’t exactly a neutral player in this discussion. Starting with his PhD at Robert Bosch and later as lead developer of Coventor’s Architect and MEMS+ software for MEMS design, he has tried to prove that a wide variety of MEMS designs can built up from a library of basic building blocks. No, there is not a single building block such as a transistor in MEMS. But yes, MEMS devices can be assembled from a set of building blocks such as plates, beams, comb drives, electrodes and anchors.

Any ally in the battle was Peter Merz from X-Fab. Foundries hate freehand layouts. Foundries have difficulties with checking designs rules on “freehand” layouts. It’s exceedingly difficult to set up automated DRC’s which in fact are written for MEMS building blocks. Hmm, this starts to sound like analog again!

But Coventor and X-Fab have been working together on all this, and along with Cadence have “conspired” to announce to create the world’s first MEMS-specific design automation enablement based on PDKs. As close to Lego as anyone has presently managed.

Gunar’s blog entry is on the Coventor website here.


Don’t Miss Mentor Graphics U2U San Jose, April 21, 2015

Don’t Miss Mentor Graphics U2U San Jose, April 21, 2015
by Beth Martin on 04-16-2015 at 10:00 pm

Mentor Graphics’ User2User conference will be held next week on April 21[SUP]st[/SUP] at the San Jose DoubleTree Hotel. This one-day, free conference is the perfect opportunity to learn, network, and share with other Mentor Graphics users.

The day starts off with back-to-back keynotes that examine different aspects of the hot topic of the Internet of Things. Wally Rhines Chairman & CEO, Mentor Graphics will address the pressing issue of Secure Silicon and how it is necessary to enable the Internet of Things. Karim Arabi, VP of Engineering, Qualcomm, will then explore the market trends and technologies driving the Internet of Things. Daniel Nenni recently discussed their keynotes here.

There’s a big new addition to User2User this year: the first-ever Emulation track, which includes an overview session from Mentor, and customer presentations from Altera, ARM, Marvell, and Soft Machines on their use of the Veloce Emulation platform. So if you’re a Veloce user, make sure you check out these sessions.

There is, of course, a free lunch at noon. You have the choice to also attend a special lunch session in the Silicon Test & Yield Analysis track. While you nourish your body, Broadcom presenter Kamlesh Pandey will feed your mind with his story of how they reduced ASIC test costs using Mentor’s new EDT test point insertion technology. Other Test presentations cover the use of IJTAG (IEEE 1687) at Cisco, finding memory failures at Broadcom, and what to do with failed EDT chain patterns at Microsoft.

I’ve written previously about IJTAG, a newly ratified standard for IP integration, access, and control that should be sweeping the nation any time now. I look forward to seeing how Cisco actually adopted and validated IJTAG and what benefits they saw.

The Broadcom presentation should be a big draw because in it, Amar Guettaf, the Technical Director of the Operations Group at Broadcom, talks about how to find the root cause of memory failures. He introduces a new bitmapping flow that apparently has dramatically simplified the MBIST bitmapping process and reduced the ATE development cycle. The last Test track presentation of the day is from Jeff Hung, Senior DFT engineer at Microsoft. He shares how they solved a situation in which scan patterns were working on ATE, but EDT chain patterns were consistently failing. All the Test track presentations are described here. Because the Test group at Mentor is full of really nice people who like to have fun, they are also offering some special prizes for their audience that you can win through a drawing.

After lunch, a panel of key executives from eSilicon, GLOBALFOUNDRIES, Mentor Graphics, TSMC, and STMicroelectronics will discuss “The Changing Foundry Landscape: Trends and Challenges.” Moderated by Semiwiki’s own Daniel Nenni, this panel of experts will discuss SoC trends, challenges, and new applications that will drive future generations of semiconductor design and manufacturing.

Along with the keynotes and industry panel, the day is packed with a full agenda of 9 major session tracks, led by top industry customers and Mentor’s top technologists. The tracks cover everything from AMS Verification (described eloquently by Dan Nenni in this blog), Calibre, Emulation, and IC Design Implementation, to PCB, Silicon Test Solutions, Thermal Simulation and Measurement and Verification.

Other activities in this full day include:

Usability Lab
Interactive sessions let you give Mentor direct feedback on product improvements.

Demo Booths
Mentor Technical staff will be available outside the session rooms to answer all of your questions.

Networking
This is your chance to network with your peers on future technology and strategy. See the snapshot to the right for an idea of what the happy hour is usually like after a Mentor U2U.

You can download the full User2User agenda hereand register for FREE here.


Intel Inline with reduced expectations-2015 flat to down-Slashing Capex

Intel Inline with reduced expectations-2015 flat to down-Slashing Capex
by Robert Maire on 04-16-2015 at 4:00 pm

Intel Inline with lowered numbers- 2015 Revs to be Flat…
Capex Slashed by 13% to $8.7B- 10nm at risk???
Mortgaging the future???
Is the foundry business dead???
Desperately seeking growth!!!

Intel Inline…

Intel reported revenues of $12.8B and EPS of $0.41 in line with downward revised estimates after chopping $1B out of Q1 previous expectations. Client computing was the main culprit for the hit down 16% Q/Q and down 8% Y/Y. Data center was down 10 sequentially but up 19% Y/Y.

The company sounds like it is hoping for some upside from the summer rollout of Windows 10 but we wouldn’t hold our breath waiting for positive impact from that introduction as we think the XP upgrade is well played out already. Guidance is for revenues to be flattish for 2015 with an obvious bias to the downside. Gross margin will be good at 61% obviously helped by reduced spending. We didn’t hear much positive hope for desktop PC’s other than the several hopefully comments about Windows 10.

We also didn’t hear a lot about tablets or mobile as those numbers are now buried in the financial results where they can’t be as easily picked apart.

No comment on M&A but lots of questions…
As expected there was no comment on the Altera rumors but it is clear from the poor momentum of the PC business that Intel has to look elsewhere for growth

Capex Cut is ominous…

Cutting capex down to $8.7B is very ominous from our perspective. Obviously Intel can help make its earnings numbers and get EPS growth in a flat revenue environment by cutting expenses of which Capex is the major number. Intel had already fallen to the number three position behind Samsung and TSMC and looks to be falling to a distant third place very quickly.

The company made the normal excuses on reuse and moving 22nm capacity to 14nm but we don’t buy that as the full reason. We find it hard to believe that Intel is that much better at reuse and efficiency than TSMC and Samsung. Capex is being cut because business is not that good. Intel did comment that yields and ramp of 14nm was better than planned but its been a long time and it should be very good by now.

You don’t get to be a leading player without spending a lot of money and we are very concerned about the company mortgaging its future to make near term earnings numbers to satisfy the street. We think this is potentially short sighted as it will help near term numbers but put in jeopardy Intel’s technology dominance that made the company what it is today. We hope we are wrong.

10nm at risk???
Though Intel refused to comment about 10nm timing it is clear from everyone in the industry that they have pushedout spending and its impossible to push out spending and reducing capex as much as they have without slowing their 10nm schedule. Our guess is that we are in the range of a 9 month to one year delay from what it could have been. 10nm was supposed to be in Israel and though Intel talked about 10nm spending in the second half of 2015 we haven’t heard a lot about it in the field.

Can TSMC catch Intel at 10nm???
Given TSMC’s increasing capex coupled with their stated aggressive 10nm plans it feels as if TSMC has a real opportunity to catch up with Intel at 10nm. There is likely a continuing downward bias in capex at Intel and its likely that management will be looking at ways to cut or delay spending to support EPS.

Foundry noticeably absent from call….
You wouldn’t know that Intel was allegedly in the foundry business from the conference call as there was zero mention of it. If TSMC does either catch Intel at 10nm or come close then the reason for a fabless customer to use Intel foundry services would fall to below zero as the only reason we can think of is their technology lead. If they lose that then foundry is officially out of business as its more expensive than TSMC and Intel is harder to work with.

Intel cutting 22nm capacity…
Intel did comment that 22nm capacity would be cut as fab capacity transitioned to 14nm. This supports to support the weaker demand environment as Intel is usually able to milk returns out of older fabs for a longer period of time. To be cutting 22nm capacity is a sign that 14nm is good but also a sign that demand is not strong enough to keep it pumping out devices.

Equipment companies likely Whacked…
The Intel capex news while not unexpected is probably a lot worse than most bullish analysts and equipment company management were hoping for. The Intel cut obviously adds to the already strong industry headwinds we have been talking about. At this point the equipment industry is standing on the one leg of memory spending as foundry and Intel (the other leg) aren’t supporting the weight. If memory spending gets more wobbly we are going to topple over.

Beware of Intel exposure…
Equipment companies that rely on Intel are obviously getting hurt but its not like anyone was expecting an increase. If Intel’s business is just a proportional share of an equipment company’s business its still hard to figure out whats going to increase to make up for the Intel shortfall. We are now potentially looking at a down year for capex as a possibility….

Robert Maire
Semiconductor Advisors LLC

Also read: Moore’s Law is dead, long live Moore’s Law – part 1


Sensing Without (much) Power

Sensing Without (much) Power
by Paul McLellan on 04-16-2015 at 7:00 am

Do you have one of those step-tracker things? They seem to be one of the earliest IoT devices that are actually selling in large quantities. Smartphones are also starting to contain this sort of sensor to provide similar functionality without requiring a separate device, as are smart-watches such as the Jumpy watch for kids on the right.

Do you know what the three top things people complain about are?
[LIST=1]

  • the app is no good
  • the sensor is not accurate
  • the battery life is too short

    One of the leaders at making the guts of this sort of product is QuickLogic with their ArcticLink 3 sensor hub along with the SenseMe agorithms that take raw sensor data and turn it into step counts. It can sense taps, wrist rotation, tell the difference between walking, running, cycling and swimming. Whether the device is in your pocket or not on your person. Whether you are asleep. For all I know it can even tell what else you might be up to in bed.


    QuickLogic is not responsible for the App directly, although they do provide drivers for Android devices using their hardware. But they have a lot to do with the other two complaints.

    The current version of ArcticLink has power as low as 75uW. Since it is always on, this is important. Further, since the sensor can detect “device not on person” it can further optimize system power by, for example, turning off I2C-bus since it won’t be needed until the device is worn again. The processor inside the QuickLogic sensor hub is a microDSP-like architecture. On its own the processor is just 32uW. That is a lot lower, for comparison, than the most miserly ARM Cortex microcontrollers. In a smartphone the power consumed by the sensor hub is a small part of the overall power consumption, but in smaller devices going from 150uW to 75uW is the difference between 3 days and 7 days or even 1 month and 2 months.


    The SenseMe algorithms are more accurate than other pedometers. There are really two forms of accuracy. The first is when you are actually moving and the device is at your ear, or in your backback or strapped to your arm. How accurate is the count? The other form of accuracy is when you are not actually walking: in a car, standing still and so on. Does the device correctly notice and count zero? A lot of competing devices do badly at this and it is very noticeable since the user knows that the count should be zero. Only the most anal of users is going to count 1000 steps and see if the pedometer got the correct answer.


    Another use for the hub in a smartphone is to help with device recovery. The sensor knows that the phone has been put down and can even tell that it has been dropped. The GPS location can be sent back to enable faster recovery even if subsequently the battery runs out and the phone is not reachable. If we all start wearing smart-watches there is no reason your watch couldn’t tell you exactly where you left your phone. Insurance companies are very interested in this too, since they pay out when people lose their phones, and so they have an interest in people losing them less often.

    Quicklogic is the only solution provider with the power consumption of a custom ASIC, the flexibility of an MCU and the algorithm capabilities of a software company.

    Details on the ArcticLink 3 S2 are is here.


  • Moore’s Law is dead, long live Moore’s Law – part 1

    Moore’s Law is dead, long live Moore’s Law – part 1
    by Scotten Jones on 04-15-2015 at 10:00 pm

    April 19th is the fiftieth anniversary of Moore’s law! We thought it would be a good opportunity to reflect back on fifty years of Moore’s law, what it is, what it has meant to the industry, what the current status of the law is and what we may see in the future.

    Moore’s law
    Moore’s law is so well known that you wouldn’t think we would need to restate it, but the fact is that many people misunderstand and misstate the “law”. In Electronic Magazine on April 19, 1965 Gordon Moore wrote: “The complexity for minimum component costs has increased at a rate of roughly a factor of two per year”. This observation became known as Moore’s law.

    Cramming more components onto integrated circuits
    By Gordon E. Moore, Electronics, Volume 38, Number 8, April 19, 1965

    At the ASMC last year, after one of the talks someone got up and said Moore’s law was a technology law but has now become an economics law. As you can see from the actual “law” it has always encompassed economics. I have seen people talk about transistors per unit of horizontal area and many other versions of the “law” that aren’t in keeping with what was originally said. Even Intel’s on-line museum doesn’t accurately quote what Moore said.

    What I believe happened after Moore’s article was published is his observation became a benchmark for the industry. Companies realized that if the industry was following such a rapid integration and cost reduction path, that individual companies must follow the same path at the same rate or be left behind. This led to a kind of technological arms race that has endured to this day.

    But Moore’s laws influence was even wider than just driving manufacturing costs and pricing. Exponentially increasing component counts and exponentially decreasing costs have resulted in the system you couldn’t build last year “because it was too complex and too costly”, becoming possible, then affordable and finally widely used in just a few short years. When Moore wrote his article many of the products we take for granted today didn’t even exist, for example personal computers, tablets and cell phones to name just a few. All of these products only became possible because of integration and cost reduction.

    The resulting new products have driven incredible growth in semiconductor revenue from less than $1 billion dollars in 1960 to approximately $10 billion dollars in 1978, over $100 billion dollars in 1994 and over $350 billion dollars in 2014. When Moore’s law finally ends it has huge implications for the entire semiconductor industry and electronics industries the semiconductor industry supports.

    Moore’s law in Action
    In case you are wondering what Moore’s law has looked like for the semiconductor industry, figure 1 illustrates the price trends for a variety of products.

    Figure 1 includes several data series:
    [LIST=1]

  • Worldwide prices for one million transistors. This is calculated by taking the worldwide semiconductor revenue and dividing it by the estimated transistors produced per year.
  • Intel price per million instructions per second of microprocessor power. This is likely the measure on the graph subject to the most error in interpretation. For many years we have estimated the processing power of Intel processes in millions of instruction per second (MIPS). There was time when Intel directly reported this number but then over time we have had to correlate against a series of new benchmarks. In the last few years we have abandoned this effort but there is still a lot of historical data. The price we pick each year is the processor with the lowest price per MIP based on Intel’s published price list.
  • Price per megabit of memory for DRAM and NAND Flash. This is simply the worldwide revenue for each memory type divided by the worldwide bits produced.
  • A trend line with a 35% per year reduction in price.

    Figure 1. Price per function trend. Source, IC Knowledge.

    As you can see from the figure, since 1980 transistors, DRAM and Intel processor prices have all followed the 35% per year price reduction closely meaning prices have actually dropped more than in half every two years! It is hard to see in the figure but for 2013 DRAM prices per megabit actually went up and then in 2014 returned to roughly the same places they were in 2012. Although this is reflective of improved pricing power for the DRAM manufacturers, it also coincides with issues in continued DRAM cost reduction that will be discussed in a later article in this series.

    It is also interesting to note that initially NAND Flash prices fell much faster than the 35% reduction seen for other products, although since around 2008 NAND prices reductions have begun to moderate and more closely follow the 35% trend. This graph and analysis are based on price trends but what about manufacturing costs? Price and manufacturing costs are related by the following:

    Price = Manufacturing Cost + Gross Margin

    Examining our data on gross margins we see flat to slightly increased gross margin over the last several decades meaning that manufacturing cost must be declining at least as fast as prices. In the next installment of this series we will examine manufacturing and see how this incredible cost reduction has been accomplished. We will then examine the history, current status and future prospects of Moore’s law for DRAM, Flash and Logic.

    Also read:
    Moore’s Law is dead, long live Moore’s Law – part 2
    Moore’s Law is dead, long live Moore’s Law – part 3

    Moore’s Law is dead, long live Moore’s Law – part 4
    Moore’s Law is dead, long live Moore’s Law – part 5


  • Nokia on Top of the World, Again

    Nokia on Top of the World, Again
    by Majeed Ahmad on 04-15-2015 at 4:00 pm

    Nokia is no more a mobile phone dynamo, but it’s now the world’s largest telecom equipment supplier ahead of Ericsson AB and Huawei Technologies. Nokia is buying Alcatel-Lucent for $16.6 billion and the new global networking behemoth created as a result of this mega-merger—called Nokia Corp.—will be headquartered in Finland.

    The second edition of my book “Nokia’s Smartphone Problem” released in November 2014 has a full chapter dedicated to the making of new Nokia from the ashes of a mobile phone giant. The book argues that the Finish handset titan had started taking its wireless network infrastructure business far more seriously after Nokia found itself flat-footed in the post-iPhone mobile era.


    Nokia’s Smartphone Problem chronicles fall in smartphones and rise in LTE infrastructure

    Nokia’s bid to acquire Alcatel-Lucent is merely the continuation of the journey that it started with the creation of the Nokia Siemens Networks (NSN) venture in 2006. Nokia combined its networks hardware business with that of Siemens and began to offload non-core assets like telecom consulting services and fiber-optic businesses. Initially, NSN, like most of its competitors, was operating in both wired and wireless infrastructure markets.

    Nokia’s focus on the Long-Term Evolution (LTE)-based 4G infrastructure business played nicely for NSN, and by 2012, it had moved from fourth to second place in the LTE equipment ranking. In 2013, when Nokia sold its mobile handset business to Microsoft, the Finnish firm also paid Siemens $2.21 billion to gain full control of the NSN venture. Now the acronym NSN stood for Nokia Solutions and Networks instead of Nokia Siemens Networks. Eventually, it settled on Nokia Networks.

    There was no doubt left in the early 2010s where Nokia was heading. The ‘New Nokia’ was inevitably about mobile infrastructure business. And the fact that Nokia had brought NSN chief Rajeev Suri to head the remaining Finnish company further cemented the notion that the New Nokia saw its future in the wireless equipment business. Fast forward to 2015, the Finnish mobile firm has completed its make-over by gobbling up a competitor of the size and scale of Alcatel-Lucent.


    Nokia buys Alcatel-Lucent for $16.6 billion
    (Image: Reuters)

    The book “Nokia’s Smartphone Problem” has made the case that the Finish electronics giant could reinvent itself like Apple and IBM. Nokia has a history of successfully adapting to market shifts, the book argued, and that Nokia’s wireless gear business could make up for its smartphone debacle. “Things could change rapidly in the technology world … and the wireless market is still a wide-open field,” the book concluded.

    Majeed Ahmad is the author of Nokia’s Smartphone Problem: The End of an Icon? The book is available in both paperback and e-book formats.


    IoT Security: Your Refrigerator Attacks!

    IoT Security: Your Refrigerator Attacks!
    by Paul McLellan on 04-15-2015 at 7:00 am

    Every time I see a presentation on IoT the forecast for the number of devices in 2020 seems to go up by a few billion. But behind the hype there are clearly going to be a large number of devices on (and even in) our bodies, our homes and cars. Not to mention in factories and workplaces. IoT devices cover a wide spectrum. Realtors like to expand desirable neighborhoods as much as they can to include whatever property they have to sell, so areas like San Jose’s Rose Garden or San Francisco’s Noe Valleygradually grow. In the same way, marketers like to jam everything they can into the IoT category even though we previously had perfectly good categories like automotive or medical. Two things, though, seem to be common to almost every IoT application: low power and security.

    Proofpoint found a wonderful example of security problems with IoT:What startled Proofpoint researchers, though, is the fact that 25% of the messages didn’t originate from the usual suspects (i.e., laptops, desktops, or smartphones). Instead, they came from connected devices, such as home-networking routers, televisions—and at least one refrigerator.

    Perhaps more worrying than being spammed by our refrigerators is infiltration of our IoT devices. If we are going to have self-driving cars we want to make sure that we decide where our car goes. If we are going to have medical devices that can, say, adjust insulin to match blood sugar levels then we want to be sure that nobody else can take control. This is not just a theoretical issue. At the end of last year, a steel mill in Germany was hacked causing “massive damage” when a blast furnace could not be shut down.

    It is increasingly clear that security requires a mixture of hardware and software. The heart of any security scheme is software algorithms along with something secret, typically encryption keys. These need to be kept in the hardware of the device so that:

    • the keys cannot be read by examination of the hardware
    • the keys are not lost when the device is powered off
    • the manufacturing cost of the key storage is minimized

    In practice this means using some form of embedded non-volatile memory (eNVM). There are a number of different eNVM technologies commercially available, with different tradeoffs with respect to cost, programmability, compatibility with process technology and so on:

    Keys are typically programmed into the device once when it is manufactured (or at most a few times over the life of the device). Antifuse one-time programmable (OTP) memory is a good match for the above requirements. It does not require a special manufacturing process like flash, it cannot be read even using expensive equipment like electron microscopes, and it is, by definition, non-volatile. It is nearly impossible to determine which bits are programmed because it is difficult to locate the oxide breakdown using chemical etching or mechanical polishing and by looking at a cross-section or top view. Kilopass’s XPM OTP memories are security certified not just for commercial use but also military. It could not be successfully attacked by either passive or invasive approaches:

    OTP memory provides best-in-class security, can be manufactured in a normal process without extra mask steps, and is low-power. In short, a perfect match for IoT.

    The Kilopass product page is here.


    Will your next SoC fail because of power noise integrity in IP blocks?

    Will your next SoC fail because of power noise integrity in IP blocks?
    by Daniel Payne on 04-14-2015 at 5:00 pm

    By the time that your SoC comes back from the fab and you plugin it into a socket on a board for testing, it’s a little late in the cycle to start thinking about reliability concerns like: dynamic voltage drop, noise coupling, EM (Electro-Migration), self-heating, thermal analysis and ESD (Electro-Static Discharge). They say that an ounce of prevention is worth a pound of cure, and that maxim is quite true when it comes to power noise integrity issues for our SoC designs filled up with re-used IP blocks and subsystems.

    You could take a detailed, transistor-level approach of using a SPICE circuit simulator during the design and layout phases to measure the effects of power integrity, except that would mean you have to wait until your design has a clean LVS netlist and all of the IC layout is completed, which is just too late during the design cycle. There is a more elegant approach that uses reliability analysis throughout the entire design process, and ANSYS is one EDA vendor with tools and years of experience in this domain.

    Early Grid Weakness Analysis

    As soon as your IP block has a GDS II layout, then you can run an early Power and Ground (PG) grid analysis to help pinpoint any areas of the IC layout that have excessive resistance and high current drive. Feedback from this analysis allows the layout designer to start fixing the PG grid at the earliest point in the design.


    IP Power Integrity Sign-off Coverage

    Static IR Drop Analysis
    As an IP block is powered up current starts to flow in the network, eventually reaching power and ground nets. The resistance of the PG nets multiplied by the current flowing creates a voltage drop, as ohms law states: V = I*R. When VDD levels lower, so does the noise margin, so it’s important to analyze each IP block with a static IR drop analysis, then inspect the grid for any hotspots identified, fixing the issue by adding vias, contacts or widening PG nets to lower the resistance.

    Dynamic Voltage Drop Analysis
    The static IR drop analysis doesn’t take into account any of the dynamic switching nature of all circuits, so a dynamic voltage drop analysis adds further reliability coverage under switching conditions. Feedback from such an analysis helps the designer to add or even reduce metal straps, vias, contacts and interconnect widths.

    Related – Noise & Reliability of FinFET Designs Need Smart & Proven Methodologies – Success Stories

    Substrate Noise Analysis
    Digital circuits that switch simultaneously can actually inject currents into the substrate, affecting the electrical performance of nearby, sensitive analog circuits, creating computational errors and degrading performance. With substrate noise analysis you can get an idea of where the noise is coming from, and how effective your layout isolation techniques are. Running this type of analysis during assembly of IP blocks will ensure that each AMS block is well isolated from noise sources.

    Related – How Early Do You Analyze Substrate Noise in SoC Design?

    EM Analysis
    There is a type of failure in aluminum or copper interconnect where the current density becomes so great that the atomic structure of the interconnect becomes altered enough to literally thin or narrow the interconnect, thereby greatly increasing the resistance and reliability of the interconnect. An increased current in the interconnect causes localized heating. With EM analysis the designer can see if their IP is over-design or under-designed for these high current effects. Both PG and signal interconnect should be run through EM analysis throughout the design process.


    EM Sign-off Flow

    Thermal Analysis

    With new transistor technology like FinFET the current drive is higher than planar devices per unit area which quickly leads to an increased thermal impact, even wires and vias can fail under high temperatures after enough cycles. You will want a thermal analysis tool capable of computing the actual thermal gradient on each IP block and then recalculate EM limits on the wires and vias.


    Thermal Integrity Coverage

    Related – FinFET Designs Need Early Reliability Analysis

    ESD Analysis
    It used to be that just IO pads needed ESD protection and analysis, however at the 65 nm and lower nodes our IP blocks need ESD checks to avert device breakdown, signals melting and cross-domain issues. DRC checks are no longer sufficient for ESD analysis, so a simulation-based approach is more thorough and trusted. The physical factors that require ESD analysis are increased resistance for interconnect, higher current densities and decreased oxide thickness for transistors.


    ESD Integrity Coverage

    Related – SoCs More Vulnerable to ESD at Lower Nodes – Must Resolve

    The specific EDA tools offered by ANSYS to help you handle power noise integrity are called RedHawkand Totem:


    IP sign-off for power noise integrity

    By running these analysis tools on each IP block, during chip assembly and before tape-out, you can validate that your SoC will work when it comes back from the fab.


    IP Integration Validation at SoC Level

    Summary

    Power noise integrity is important to the reliable operation of your IP-based SoC designs, and by using the methodology described it will ensure that your next project works to spec without surprises.