We’ve introduced the concepts behind triple modular redundancy (TMR) before, using built-in capability in Synopsys Synplify Premier to synthesize TMR circuitry into FPGAs automatically. A recent white paper authored by Angela Sutton revisits the subject Continue reading “TMR approaches should vary by FPGA type”
IC design challenges are different at advanced nodes like 7nm, so to learn more about the topic I attended a panel luncheon at DAC sponsored by Cadence. The moderator was both funny and technically astute, quite the rare combination, so kudos to Professor Rob Rutenbar, a former Neolinear guy now at the University of Illinois. Panelists included the following people:
Continue reading “IC Designers talk about 28nm to 7nm challenges at #53DAC”
The definition of IoT systems is moving fast: the simple definition of a connected (to Internet) device in the early days is becoming more complex and also more precise. IoT or M2M devices are expected to support the full range of emerging cellular protocols such as LTE MTC Cat-1, Cat-0 or Cat-M, as well as the suite of Low Power Wide Area Network (LPWAN) standards such as Lora, SigFox and Ingenu. IoT and M2M will also interface with other IoT-related communication standards, including Wi-Fi 802.11n, PLC, 802.15.4g, ZigBee/Thread, GNSS or any emerging protocols such as NB-IoT and Wi-Fi 802.11ah.
The IoT or M2M designer has the choice between integrating CEVA-XC8, the fifth generation of the widely licensed CEVA-XC architecture, optimized for IoT communication applications as well as UE terminals, or CEVA-XC5. This sixth generation, the CEVA-XC5 is optimized for IoT communication applications, 802.11n, PLC, 802.15.4g, GNSS as well as other connectivity applications. The CEVA-XC5 delivers highly powerful vector capabilities alongside a general computation engine to supply the performance and flexibility demanded by various IoT communication applications, while meeting the strict demands for power efficiency and low cost.
Developing an IoT or M2M device implies selecting cost optimized architecture, exhibiting the highest possible performance to power ratio, leading to select CEVA-XC5 DSP core.
Designing IoT or M2M platform will lead to select the CEVA-XC8 delivering highly powerful vector capabilities alongside a general computation engine to supply the performance and flexibility demanded by various IoT communication applications.
According with Linley Gwennap, principal analyst at The Linley Group, ”designers of multi-standard, low data rate LTE-capable devices have two priorities: minimizing cost and power, and with the CEVA-XC5 and CEVA-XC8, CEVA has leveraged its industry-leading position in communications DSPs to deliver the small die size and high power efficiency that these designers need for IoT and other applications.”
Once the IoT SoC architecture has been defined, using CEVA-XC5 for a device or CEVA-XC8 for a platform (or base station), the project manager has to care about software development and make sure to use concurrent engineering to develop software and hardware in parallel to optimize the project development schedule. We know that the software development is expected to consume 2/3[SUP]rd[/SUP] of the project resources in average. The project manager has to make sure that the DSP IP core supplier will also propose a software development platform.
This is the CEVA Dragonfly reference platform, pictured below, including DSP drivers, RTOS, DSP library, C run-time library and also LTE MTC Modem and Voice & VoLTE libraries. On top of these libraries, the designer could take benefit of existing software IP, available to support a project development the same way than hardware IP.
Taking the example of satellite localization for M2M or IoT, let’s get the feedback of Eli Ariel, CEO at Galileo Satellite Navigation. “The CEVA Dragonfly reference platform delivers exceptional performance for implementing our software-based GNSS receivers in devices within a stringent power budget,” said Eli Ariel, CEO at Galileo Satellite Navigation. Our Software Receiver solution perfectly complements CEVA’s software-based approach to design flexibility and long service life MTC systems design, allowing customers to carry out performance improvements and new features in the field, including upgrading to future satellite systems.”
This comment from Nestwave CEO helps understanding how powerful can be the Dragonfly reference platform to speed up project development schedule: “Accurate positioning, both indoor and outdoor, will be a fundamental component of many M2M applications and our CellNav™ technology delivers this accuracy utilizing the existing LTE network infrastructure,” said Rabih Chrabieh, CEO of Nestwave. “Using the CEVA Dragonfly platform, customers can integrate CellNav into their MTC product designs, enabling reliable location tracking in devices that can last years in the field on a single battery.”
I suggest you to attend remotely to this webinar, describing a real design case using CEVA-XC5 and Dragonfly reference platform:
Webinar: Design a LTE-based M2M Asset Tracker SoC with CEVA, using GNSS and OTDOA
Eric Esteve from IPNEST
Synaptics is an understated innovator in the human interface and end user experience technology space, and many of the hottest smartphones, tablets and PCs have their technology inside. The Samsung Galaxy S7, Apple iPhone 6S, Dell XPS 15, Microsoft Surface and Surface Book, and HP ENVY are just a few of these products I’ve personally used. One of the interface technologies that Synaptics excels at are fingerprint sensors.
While most of the PC “swipers” are Synaptics today, the company has been working with both PC OEMs and smartphone vendors over the last two years to integrate their latest fingerprint technologies into a multitude of devices. Synaptics has also taken their finger print sensor and integrated them into their own touchpads and has the unique position of being able to get their customers to integrate both into their devices. However, the fingerprint sensor ecosystem requires a robust back-end to enable use cases outside of secure PC login and password replacement. While PCs were the first to have biometric finger “swipers”, the PC has lagged behind smartphones and tablets. I’m hopeful Synaptics latest technology can change all that.
USB Fingerprint Security Key mockup (Photo credit: Synaptics)
Synaptics has been working with partners like Microsoft and Lenovo to make computers more secure with multi-factor authentication. Biometric methods of authentication help to reduce the probably of unauthorized access and at the same time improve the user experience. We have written on that here. Synaptics’ Natural ID technology is designed to work with technologies like Windows Hello and Microsoft Passport, which already give them legitimacy in the Windows 10 world. Microsoft Windows 10 is designed to be the most secure version of Windows ever while also being the most user friendly, so native support for things like biometric authentication are logical.
Synaptics is introducing a new ultra-small form factor USB module that enables the use of their Natural ID fingerprint authentication on any PC with a USB port. This module will primarily be focused on notebooks and notebook users, but could theoretically be used for any PC. The USB dongle is a turnkey solution for OEMs, ODMs and other manufacturers to offer an inexpensive finger print sensor into PCs without one. The dongle is small enough to remain installed in any notebook USB port and with enrollment can be used to enable features like Windows Hello and Microsoft Passport.
This solution is merely a short term solution until more PC manufacturers like Dell, Lenovo, HP and Apple figure out that fingerprint sensor technology has reached a point where it both improves experience and security. With nearly all flagship and mid-range phones featuring fingerprint sensors, it is almost inevitable that fingerprints will drive a lot user experiences on all platforms, not just mobile. The company’s new fingerprint reader adds yet another form factor to the other offerings that Synaptics offers their customers in order to best satisfy their form factor needs. In some notebook designs it may simply not make sense for them to integrate a fingerprint sensor into the wrist pad or touchpad quite yet but the manufacturer may still want to offer fingerprint sensing as a feature.
While I am disappointed that PC manufacturers like Apple, Dell, Lenovo and HP Inc. haven’t included enhanced fingerprint readers yet into their personal computers, I am hopeful that 2016 will be the year this happens. There are pragmatic reasons they haven’t done this, but the lack of integration is yet another example we can all point to showing PCs are behind the curve.
Synaptics is constantly pushing the industry towards new user interface experiences and improved security. With their new USB dongle, they are making it easier for anyone to implement a fingerprint sensor on their PC and integrate it with next-to-no engineering effort. There are also opportunities for people to add this dongle to their keyboards and other USB peripherals that might have a USB port. The fingerprint sensor dongle will sample in Q3 this year and mass production is expected in Q4 this year, meaning that we could see these very low profile fingerprint dongles in PCs by the end of this year.
Hype-cycles are well known phenomena to those who pay attention to high-tech markets. Since we are probably at a peak of the early IoT hype-cycle in 2016, it’s natural to expect this latest wave of technologies to evolve just like other tech categories have in the past, with the eventual market power and market value going to a few consolidated, independent, and market leading platform companies. I think making this assumption could be a big mistake, particularly when it comes to platforms for IoT enablement (connected device platforms).
In the tech industry, industry platforms get created when markets consolidate around a small number of players who deliver standardized services (the platform) that enable application development and deployment. From the perspective of a technology business, history suggests that companies with new disruptive innovations repeat a similar process to eventually achieve market-leading platform status:
[LIST=1]
You might also expect this process to result in high business valuations and industry domination by a few independent tech company winners. After all, when it comes to application enablement platforms, when markets mature, a small group of companies will usually end up dominating as their customers flock to the safe choices. Right?
Maybe not. In fact, there are good reasons why IoT technology categories may evolve very differently. Companies within previous B2B application enablement categories like app servers, BPM, databases, ERP, marketing automation, messaging, sales automation, switching, and others, grew their power (and market valuation) by consolidating services to support multiple use cases. As they consolidated, they achieved rapid revenue growth and built entrenched partner ecosystems, leveraging high switching costs, dominating market share, and controlling architectures that became widely adopted.
In essence, they helped to create and then control the complete industry value chain built around their platform.
This was the playbook that category leaders from Autodesk and Adobe to SAP and MS/Windows executed to become market dominators. Unlike with these past categories, however, the value chains for connected device applications are going to be much more difficult for technology companies to directly control. In my opinion, IoT infrastructure providers will face significantly bigger challenges in achieving the kind of mega-Gorilla status and sustained high valuations than did their predecessors.
In the rest of this post I will refer to “the customer” as the customer of a technology companywho sells IoT infrastructure. I won’t get into the details of either the IoT platform value chain or the IoT business models here, but as reference, two recent posts by Microsoft’s Mohit Agrawal nicely illustrate both: the first post, “Business Models in the Internet of Things” illustrates the IoT value chain. The second post, “Impact of Internet of Things on Business Models,” summarizes different business models that are coming into play.
IoT is about accessing, analyzing, and operating upon information retrieved from edge devices, and then using that information to create efficiencies and improvements to the performance of those edge devices (or the end customer’s experience). The value chain doesn’t just end at the customer (the organization which a tech company sells to). It extends to the customer’s customer (or the customer’s machine) at the network edge. The insights generated from the data are a major part of the value created — what edge devices are doing, when they need maintenance, how or why consumers are using them, and what might be expected to happen next.
Companies who want to expose this value require a way to optimize machine or consumer experience through the use of increasingly sophisticated analytics. This is key, because unlike the other categories I referenced previously, the companies who embark upon a connected device strategy (the customers of technology companies)are likely to see IoT infrastructure as a strategic investment into their core businesses and into their actual products and processes.
This viewpoint will make it more likely for these companies to interact with technology providers strategically, and even in a competitive way. They may need the tech companies to initially assemble solutions and build market entry. But, just like independent software companies have done in the past, as markets mature and growth eventually slows, they will will start looking for ways to design the partner tech companies out of the game, either through independent in-house development, consolidation, or acquisition. Since the application value to be unlocked requires access to the edge-point data, and because the manufacturer owns the access to that data, the real power in the IoT market is much more likely to migrate towards manufacturers rather than only to the tech companies who support them.
Consider this:
when application infrastructure such as ERP, messaging, CRM, marketing automation, and others were originally adopted by major manufacturers such as GE or Bosch, it was undoubtedly clear to them that building and controlling the software platforms internally made little sense. It was a different business than the ones which they were in; the competencies required were different, and, had they invested in trying to do it themselves they’d have gotten zero credit in the public markets for the effort. Today, however, it’s not so clear. Large corporations that plan to implement connected device strategies may see platforms that control, manage, analyze, maintain, and update devices as natural investments into their core businesses. They may also see the IoT as a way of transforming aging commodity product businesses into modern “products-as-a-service” businesses with more predictable revenue streams.
In fact, both GE and Bosch appear to have reached this conclusion. GE, for example, has made heavy investments into creating its GE Digitalgroup, and establishing its Predix IoT platform not just for use within its own divisions, but for use by other organizations and industries. It’s invested at least $ 105M into the big data and cloud PaaS company Pivotal. Bosch, recently acquired Cologne-based ProSyst Software GmbH to fuel its own IoT offerings, and says it will also make the software available to other companies. Unlike the early days of front and back office application enablement categories, manufacturers such as GE and Bosch no longer see commercial software as something far afield from their core businesses. In fact, they are investing millions into software enabled offerings of their own.
Unlike earlier technology categories, where independent, market-dominating technology providers grew as independent companies, it’s more likely for key IoT platform solutions to end-up under the eventual ownership and control of the companies who manufacture the “things” of the IoT. Their primary competitors may not be Gorilla-like independent software companies that deliver a complete solution, but consortia-driven, open source platforms.
Another key difference with IoT applications has to do with how value is created. Because connected device applications get much of their value from analytics sourced from edge device or end-user data, independent IoT platform providers can be blocked by their customers from creating and owning analytics-based applications due to the sensitive nature of the customer data required.
IoT platform providers can certainly provide tools that enable customers to create analytics — for example, descriptive analytics that report things about where edge devices are at, how they are performing, what they are doing, etc., but this level of analytics is always the first to commoditize. The larger value from a connected device strategy comes from predictive analytics — machine learning and prescriptive alerts that operationalize and optimize device performance. Creating these high-value predictive use cases requires open access to statistically significant edge device data to understand whether an application is even possible. Tech companies can get access to this sensitive customer data doing work for hire, but they are often prohibited from owning the resulting intellectual property that could enable them to build reusable and marketable business applications.
Nick Beimrecently wrote a TechCrunch article The Barbell Effect of Machine Learning, pointing out how these higher forms of analytics will likely concentrate value among those who own the data. I agree. Within the IoT market, I believe that other major manufacturers, just like GE and Bosch, will come to realize that the data harvested from their edge device installed bases, and the corresponding analytics, are the golden jewels for creating business value. And, if they are making the investment into developing this infrastructure for their own divisions, why not leverage the efforts to create the same tools for other businesses to use?
So, what about the independent software vendors such as PTC, who acquired Thingworks and Axeda, or Autodesk, who acquired SeeControl? Both companies made these acquisitions very early in the IoT life-cycle, and each of these acquired businesses now has deeper resources and an existing customer base to up-sell into. Even so, I don’t believe that industry consolidation of the type we saw with ERP, sales automation and the other B2B mega categories is likely to happen. Companies such as Autodesk and PTC will still face the value chain barriers that did not exist with PLM/PDM and CAD tools. Over time, their competition will include companies such as GE, Bosch, and probably many other industrials who can differentiate through deep domain knowledge at the high end, and at the low-end, the possibility of open-source and consortia led offerings.
If I’m even a little right about this, the eventual shake-out will produce a more fragmented industry as compared with what we’ve come to expect based upon history. Connected device players who execute a platform strategy and remain independent will find it harder to achieve real market dominance. These can still be tremendous businesses, of course, but they might not achieve the huge platform valuations that previous tech Gorillas did in their heyday. At the same time, established Fortune 100 companies who choose to invest into building connected device platforms themselves may increase their bottom lines through efficiencies and their top lines through recurring services, but the financial markets will probably just value them as better versions of themselves unless or until they eventually spin out those businesses as independently valued pure-play companies. Either way, it will be interesting to see how this market evolves.
“The Soviet Union I left behind was a dictatorship but the workplace was a democracy; America may be free but the workplace is a dictatorship” said Len Erlikh after I hired him at First Boston (now Credit Suisse First Boston) in 1986. Being of the Jewish faith, he had fled the U.S.S.R.’s religious persecution.
Erlikh’s words have always stuck with me. It is true that in capitalism, the workplace is mostly autocratic. You do what you are told and don’t have any say in the company’s strategy and operations. The Soviet collectives allowed much greater worker participation — and that is probably why they failed.
Business leadership is not a popularity contest; the best companies are run by enlightened dictators.
CEOs must listen very carefully to their employees but they have to do what is best for the company, employees, and shareholders. They have to make tough decisions and take responsibility when things go wrong. They expect that once the decision is made, everyone will comply — whether the decision was good or bad. The best leaders share the credit when they achieve success and take all the blame when things go wrong.
I know that dictatorship doesn’t sound nice but it is what business leadership entails. People love to follow strong leaders. They want to be led by people with vision, conviction and good values. They may not agree with everything the leader decides, but as long as ethical lines are not being crossed, employees will follow directions, work hard, and be loyal.
Look at some of the most successful business leaders:
Autocratic leadership only works until it doesn’t work, however. And then everything goes wrong; entire companies collapse. Autocratic CEOs often become the bottlenecks in decision making because everything has to be approved by them. And they cause employees to stop taking risks because they become fearful of making the wrong decision. These CEOs start believing their own press and lose touch with what made them successful. If you look at any list of defunct companies that were household names, you will find misguided autocrats at their helm.
There needs to be a balance between strong leadership, autonomy, and empowerment of employees. And leaders need to step aside when they have peaked as Cisco CEO John Chambers did last year. He too was an autocrat who said to the New York Times “I’m a command-and-control person. I like being able to say turn right, and we truly have 67,000 people turn right.”
Chambers realized technology was making it possible for leaders like him to rule in a better way, with more collaboration and teamwork. He said in 2009, “If you had told me I’d be video blogging and blogging, I would have said, no way. And yet our 20-somethings in the company really pushed me to use that more.”
The job of manager today is to lead, articulate goals, inspire, motivate, and enable. CEOs must facilitate rather than control as well as listen and communicate. With technology, they can get input from every part of the company and explain the unpopular decisions. Through email, internal social media, and idea exchanges, companies can have everyone participate in problem solving.
Employee engagement can be done on small decisions as well as big ones. In February, IBM made a big decision, to revamp its global performance evaluation system, by crowdsourcing the solution. It explained the deficiencies of its old system to its 380,000 employees in 170 countries through its internal social media platform and asked them to suggest solutions. Based on the 2000 comments it received, IBM ended yearly reviews and replaced these with a system for shorter-term goals and quarterly feedback. These are the types of structural changes that are needed in today’s era of exponential technologies — in which a year is a lifetime and shifts in strategy are needed every few months.
Leaders can be dictatorial yet inspire and motivate if they listen and communicate effectively — and honestly. To survive the disruptions that technologies will cause in practically industry, companies will need enlightened dictators who have a heart.
For more, follow me on Twitter: @wadhwa and visit my website: www.wadhwa.com.
The media is really having a field day on this one so I think it deserves further discussion. The rumor is that Intel has won the modem socket in the iPhone 7. The same rumor was circulating about Intel winning the modem socket for the iPhone 6 and the iPhone 5e so it really has reached urban legend status. The question I have is why does anybody really care? The modem in question was not designed by Intel, it is not manufactured by Intel, and it does not guarantee Intel another iPhone modem socket, so seriously, what is the big deal here?
First a little background: QCOM has supplied iPhone modems for the past few years. My iPhone 5, 5s and 6 has a 28nm QCOM modem. The iPhone 6s has a 20nm QCOM modem and right now QCOM is shipping a 14nm modem which is inside the latest and greatest Samsung smartphone (S7). I’m not going to get into the modem speeds and feeds debate because it is mostly “benchmarking magic.” Seriously, even Harry Potter would be impressed by some of those benchmark claims and we are all carrier speed (Verizon/AT&T) limited anyway so it doesn’t really matter.
According to Bloomberg:
“Apple Inc.’s next iPhone will use modems from Intel Corp., replacing Qualcomm Inc. chips in some versions of the new handset, a move by the world’s most-valuable public company to diversify its supplier base.
Apple has chosen Intel modem chips for the iPhone used on AT&T Inc.’s U.S. network and some other versions of the smartphone for overseas markets, said people familiar with the matter. IPhones on Verizon Communications Inc.’s network will stick with parts from Qualcomm, which is the only provider of the main communications component of current versions of Apple’s flagship product. Crucially for Qualcomm, iPhones sold in China will work on Qualcomm chips, said the people, who asked not to be identified because Apple hasn’t made its plans public.”
So they are telling us that AT&T iPhones will have a 28nm Intel modem while Verizon iPhones will have a QCOM 14nm modem? Would Apple really do this? Is Apple really going to pair a leading edge FinFET SoC with a 28nm planar modem? Let’s not forget about the infamous Battery Gate issue between the Samsung and TSMC versions of the Apple A9 SoC. Some clever fellow even made an app to tell us which chip was inside. I betcha Apple did not see that one coming:
Or will Apple use 28nm modems in all of their iPhones while their fiercest competitor (Samsung) uses QCOMs 14nm modem which is two process generations ahead of Intel?
And how much is Intel actually going to make on a modem chip that they did not design or manufacture? Certainly not enough profit to be noticeable, especially if there is Contra Revenue involved.
In the past I would have chocked this rumor up to media insanity but after Apple split the A9 between Samsung and TSMC I can only shake my head. Apple does punish its suppliers so maybe this is a warning to long time modem partner QCOM. But, by punishing QCOM Apple is rewarding TSM since TSMC manufactures the Intel modem in question so it’s all good for the fabless semiconductor ecosystem.
And just when I thought the semiconductor industry couldn’t be more interesting!
Advanced Micro Devices has already told us that 2016 was going to be the year of graphics, but the reality is that they also have a lot going on in their CPU and APU division as well. In fact, in addition to Advanced Micro Devices’s newly announced 7th Generation APUs in 2016, the company is also expected to launch their new Zen CPU cores which are being eagerly awaited by many in the industry. Zen won’t be in high-volume until 2017, so tweaks to the current generation is important. AMD’s 7th generation APUs show that even though we are constantly paying attention to architectural improvements when it comes to CPUs, there are still plenty of platform improvements and tweaks that can be made to squeeze additional performance out of it. It may not get the headlines, but it drives revenue, something the company needs now.
AMD 7th Gen Die Shot (Photo credit: AMD)
Beneficial performance and power upgrades for PCs
The new 7th generation of APUs is similar to the previous generation that it still uses the 28nm fab process and still has four Excavator CPU cores and AMD Radeon GCN GPU cores. However, this new “Bristol Ridge” codenamed part is designed to be much faster than the previous Carrizo APU. Advanced Micro Devices has upgraded the GPU in the Bristol Ridge APU to allow for performance improvements of up to 37% according to AMD’s own numbers. We have not received parts and benchmarked them, but will if we do receive them. The new generation of AMD’s APUs also accomplish multimedia tasks at up to 12% lower power than the previous generation according to Advanced Micro Devices’ numbers.
Many, smaller improvements add up to some big improvements
These performance improvements weren’t a result of a process node shrink but rather a series of smaller improvements that could amount to something like a process node shrink. Advanced Micro Devices did, however, experience a small benefit to frequency and power consumption due to a process node improvement. Advanced Micro Devices has also implemented what are called “Shadow P-states” to achieve the best possible clock speeds for a given chip. Part of what enables the higher clock speeds is Advanced Micro Devices’s implementation of Adaptive Voltage and Frequency Scaling (AVFS).
Advanced Micro Devices also implemented a reliability tracker on the “Bristol Ridge” APUs that allows the chip to operate at its optimal frequency to maximize its life and reliability while also maximizing clock speed. To further maximize performance, Advanced Micro Devices implemented a skin temperature aware power management system that allows Advanced Micro Devices’s APU to utilize the entire device as a thermal sink in order to allow more sustained performance. A special controller was designed to model the thermal capacitance of the given platform and determine the appropriate thermal headroom to allow for certain clock speeds and durations. This feature was originally implemented in Advanced Micro Devices’s smaller Mullins APU, but this is the first time that AMD is implementing it on a larger APU for a bigger market.
All of these improvements are in addition to the upgraded video decoder, faster DDR4 memory support and faster and more GPU cores. This includes support for Google’s VP9 video codec and 4K HEVC which are both necessary for any future video playback capabilities. The new product lineup will range from a 15W Advanced Micro Devices E2-9010 APU with a 2.2 GHz max frequency all the way up to a 34W Advanced Micro Devices FX 9830P with a maximum clock of 3.7 GHz. Advanced Micro Devices now has products that fill many of the different performance and TDP needs for the majority of the market with these parts. One product that Advanced Micro Devices is showcasing with their new Bristol Ridge APUs is HP’s ENVY x360 which is premium convertible notebook that is available now. It says a lot that AMD’s new APU was selected for HP’s premium device as HP is on the rise in this space. I wrote about that here.
HP ENVY X360 (Photo credit: HP Inc.)
An interesting remainder of 2016 for AMD
Advanced Micro Devices is going to have an interesting 2016. The company is clearly working very hard to recapture market share on both CPUs and GPUs and revive their APU line of products. They are clearly making some meaningful improvements and trying to innovate where they can while still staying on the same process node.
Ex chip-jockeys like me are very impressed with what AMD has been able to do with the legacy node
Bring on Zen
Once Advanced Micro Devices switches to 14nm FinFET with their Zen CPUs we can finally start to expect Zen APUs with Polaris GPUs for the graphics. However, in Anshel Sag’s conversations with Advanced Micro Devices executives at Computex, it was indicated that this is unlikely to happen until 2017. Until then, parts like Bristol Ridge are going to have to be the best of what Advanced Micro Devices has to offer with all of the optimizations and tweaks to make it a better and smarter chip than the previous generation.
AMD CEO Lisa Su gives a Zen sneak peak in Taipei (Credit: AMD webcast)
For CPUs, it’s only up from here for AMD. Bring on Zen. And at AMD’s Computex event in Taipei, CEO Lisa Su re-confirmed the 40% IPC improvement with Zen. I don’t know how this is possible, but if it plays out across most workloads, PC and server CPUs just got infinitely more exciting in 2017.
“I AM ZEN” proclaimed AMD at Computex 2016 (Credit: AMD webcast)
Last week at the #53DAC conference there was a lot of excitement in the air about custom IC design, especially at the luncheon that I attended on Tuesday from Synopsys where they had customers like STMicroelectronics, GSI Technology, Samsung Foundry and the Synopsys IP group talk about their experiences using the new Custom Compiler tool. The luncheon that they hosted was in the Hilton, and had the best food service at DAC this year, plus the room was packed with interested engineers.
STMicroelectronics
Atul Bhargava started out by explaining that his company helps make products for the smart city, smart home, smart driving and the IoT. Their primary technology is FDSOI and so they create their own libraries, foundation cells, I/O, memories and AMS designs. When they did a survey of their own IC designers it was discovered that their actual time was spent with 30% on transistor and cell placement, 25% in routing and 45% in validation (DRC, LVS, EM, IR drop, reliability).
With the Customer Compiler tool they like using the symbolic editing in Custom Compiler, and how it helps automate digital layout devices. Analog designers can quickly place dummy devices and use common centroid topology automation. Even their analog layout patterns can be saved for future reuse. EM analysis is a quick process in Custom Compiler, even pushing it to an earlier part of the design flow right after schematic capture where they can start to plugin in currents for Peak, Average and RMS, then decide how many vias are required for a detailed layout.
The look and feel of Customer Compiler is very much like a modern web browser, so it’s intuitive and easy to learn.
This group has been able to take an iPDK from 28nm PDSOI and quickly migrate it, plus all of the required IP migrated. Even the SRAM and standard cell groups are now using Custom Compiler.
GSI Technology
Up second was Randy You, and his company was founded in 1995 with offices in both Sunnyvale and Taiwan, growing to 140 people now. Their IC designs are differentiated from competitors by offering higher performance and lower power consumption. On a recent 16nm project they started using the Custom Compiler tools and quickly noticed three areas of design challenge: design rule complexity, EM/IR analysis and balanced net routing.
With the Custom Compiler layout tool they use the Dynamic Rule Distance (DRD) feature to show layout designers if there are any design rule issues as they do custom layout, it even works with double patterning. A full DRC is still required after using the DRD feature, but using DRD reduces the number of full DRC runs required.
In their EM/IR design flow they can start with schematics, run circuit simulations to create currents, then the current values are used to quickly recommend metal widths required for layout, even before running a detailed EM/IR analysis, reducing layout/analysis iterations and saving time.
Balanced nets can now be auto-routed instead of manual routing, the clock tree can be auto routed to all I/O block.
Samsung Foundry
Next we heard from Bonhyuck Koo about their iPDK development group that does FinFET design, uses multi patterning, and combats design rule complexity. On their 14nm FinFET process the Customer Compiler tools were used in a color-aware design flow. They can even do density checks with coloring and have automated coloring.
For DRC and LVS checks they use an in-design approach, instead of solely relying on a batch approach which causes too many iterations to reach a clean layout. Doing early EM checks also enables designers to meet reliability requirements earlier in the tool flow, instead of waiting too late in the flow. The 10nm iPDK is coming next month in July, so stay tuned for the formal announcement.
Synopsys IP Group
Did you know that Synopsys has about 2,400 engineers doing soft IP design and 1,300 AMS IP designers? They use their own Custom Compiler tools for all of their work and have done test chips taped out for Samsung 14 LPE, 14LPP, TSMC 16FFP and others. Synopsys has been doing FinFET designs since 2012. Their custom IC challenges are: DRC complexity, EM/IR analysis, density, performance, coloring, segmentation, and the annoying fact that fin pitch doesn’t equal metal pitch.
Working with the foundries they get early PDK layout rules and start making their new cells with an eye towards allowing changes as the rules get updated. The EM assistant helps reduce iterations and closure for reliability concerns. A video comparison showed the same custom IC layout task with and without all of the Custom Compiler automation features, and the net result was a whopping 7X improvement.
Summary
Users have some real choices today when it comes to custom IC layout design tools, and Synopsys appears to have reached critical mass with their Customer Compiler tools. It’s worth taking a look at their new tools to see how they compare with your existing design flow productivity.
Of late, it has become painfully obvious that the value of electronics is in the system. And since systems demand continuing improvement, increasing performance and decreasing cost (once partially guaranteed by semiconductor process advances) is now sought through algorithm advances – witness the Google TPU and custom fabrics for high-performance server designs. But designing algorithms in RTL would be a masochistic exercise. The right place to do this is in software, whether C/C++, MatLab or similar platforms.
Another important change is in increasing use of software-based verification. Design verification is running on emulators and prototype platforms on more of the software stack, in order to provide confidence and coverage across a range of use models. These factors together mean significant components of the design and overall SoC verification are increasingly centered on software rather than RTL. The Mentor Calypto group have recognized this shift and are building a solution to address both design and verification together at high-level design (with appropriate re-verification at RTL).
The synthesis part of the story is or should be well known by now. Catapult may well be the most capable of the commercial HLS solutions because it can synthesize from both SystemC and C++. It also allows for untimed, loosely-timed or cycle-accurate models, providing the ability to use timing-based constructs where needed but to expand to full C++ for complex algorithm development. And naturally algorithm design, experimentation and synthesis at this level is more productive than at RTL.
The verification part of the story is where this gets really interesting. If you know Calypto, you know they also have high-level property checking (non-temporal today). This is now embedded in the flow, meaning you can check for things like array-bounds errors, incomplete cases statements and user-defined assertions while you are still working with the software model. And since dynamic verification on a software model running software-based tests runs 1000’s of times faster than any RTL-based testbench could hope to run, this is a much better place than RTL to do intensive testing across many realistic use-cases.
The Catapult verification flow also automatically runs verification on the generated RTL using the same high-level testbench and compares results between the two simulations, all tightly coupled to Mentor verification solutions. Further, C-based assertions and cover statementrs will be converted to (synthesizable) OVL or PSL equivalents in the generated RTL so they can be checked in RTL-based verification.
This tight coupling gets you pretty close to RTL verification signoff without needing to lift a finger in RTL testbench generation and debug. Pretty close but not perfect and Mentor freely admitted as much to me. Control logic added during synthesis isn’t tested by the high-level test bench, so isn’t tested in RTL verification either. Verification engineers have to add their own tests for these components, e.g. for stall and reset tests and unreachability, a task which seems not too onerous today, at least based on customer experiences. I wouldn’t be surprised to see Mentor patch some of these automation holes in future releases.
All nice, but customer results are the real test. NVIDIA, on a 10M gate video decoder for Tegra X1, were able to improve design productivity by 50% and cut verification cost by 80% (from an estimated 1000 CPUs for 3 months to 14 CPUs for 2 weeks). In a later spec revision, they were able to re-optimize the IP from 20nm/500Mhz to 28nm/800Mhz in 3 days!
Google built a similar video decoder. They were able to get from start of design to verified RTL twice as fast as they had projected without this flow. And IP modifications were 4X faster. If Google thinks this is a good idea, might want to ponder the way you are doing it today, even if you believe you get better results. In the systems world (aka the majority of our volume market), schedule is generally more important than the highest-possible performance result.
All of which makes me wonder what role RTL will have to play in the future, at least at the system level. Does RTL become the new gate-level, good for emulation and ECOs, but otherwise destined to remain unseen? You can learn more about Catapult advances HERE.
