webinar IPXACT banner

The Case for Data Management Amid the Rise of IP in SoCs

The Case for Data Management Amid the Rise of IP in SoCs
by Majeed Ahmad on 09-02-2015 at 12:00 pm

In the late 1990s and early 2000s, during the adolescent days of the system-on-chip (SoC) design movement, there was a lot talk about IP and design reuse, but it was seldom put into practice. A decade later, SoC turned into a juggernaut with a tripartite alliance of chipmakers, IP suppliers and semiconductor manufacturing fabs.

The internal and external IPs became a key building block in modern SoCs due to a number of semiconductor industry developments. For a start, the leading-edge SoCs continued their journey toward smaller nodes while striving to overcome speed and power challenges. Then, there has been a proliferation of standards. Take the Internet of Things (IoT) chips, for instance, where wireless standards like Bluetooth, Wi-Fi, ZigBee have become check-box items with the integration of IP subsystems.

A vantage point look at the evolution of SoC design over the course of a decade or so shows that the addition of more functions onto a single chip led to more complexity as well as time-to-market pressures. That made SoC segment an industry in a hurry. Chipmakers are now obliged to work around tight delivery schedules in the new IP-centric SoC model, and that makes the ever-growing library of IP content too hard to handle.

As a result, a new breed of semiconductor outfits have emerged that is benefitting from the growing complexity and time-to-market bottlenecks, and subsequently, the rising amount of IP in SoCs. One such company is ClioSoft Inc., the Fremont, California–based supplier of automated solutions for hardware configuration management of SoC designs.


Data management eases handling of disparate pieces of IP in SoC

ClioSoft, founded by Srinath Anantharaman in 1997, pioneered data and IP management for chip design on the lines similar to software configuration tools like Subversion. Anantharaman could see how precious chip engineering resources were spent on the manual handling of tasks like revision control, issue tracking and other gatekeeper functions.

ClioSoft’s first product—Save Our Software or SOS—managed the front-end RTL flows and eventually became a natural fit for data management in Cadence Virtuoso-based analog/mixed-signal designs. Over the years, the SoC design ecosystem continued to evolve and so did SOS tool as it added other popular analog/mixed-signal design flows such as Mentor Graphics Pyxis and Synopsys Custom Designer and Laker.

Then, in 2013, ClioSoft announced another milestone with the integration of its SOS tool into Agilent Technologies’ Advanced Design System (ADS) to provide version control and enterprise-wide management for RF and high-speed digital designs. That year, Agilent also spun off its test and measurement business as Keysight Technologies, which now owns and operates the prevalent RF design tool ADS.

ClioSoft’s SOS design collaboration platform is also winning attention from semiconductor IP suppliers, a crucial part of the chip design ecosystem. The steadily increasing use of IP in today’s bigger and more powerful SoCs means that IP vendors need to be more astute in customizing their products for different chipmakers and foundries, and do it in a time-efficient manner.

Not surprisingly, therefore, IP vendors are now increasingly using data and IP management tools like SOS to tag their products and hence map a specific IP from one chip client to another and from one foundry process to another. The design changes are happening on multiple fronts in an often geographically scattered SoC project, where design automation tools like ClioSoft SOS can play a critical role in efficiently managing the IP labyrinth.

Also read:

ClioSoft SOS v7.0: Faster, Smarter and Stronger

Starvision and SOS, a Perfect Match

Why Design Data Management: A View from CERN


Threat Detection: How To Keep the Crown Jewels Secure

Threat Detection: How To Keep the Crown Jewels Secure
by Paul McLellan on 09-02-2015 at 7:00 am

Let’s just take it as a given that securing IP design data is critical. It’s rather like saying that it’s a good idea to have security in the Tower of London to stop the crown jewels being stolen. IP blocks are the crown jewels of an SoC company.

Data now must be secured within the collaborative teams that share that data across international boundaries. Adding to the challenge is the fact that most of the current generation of security tools are perimeter-based solutions, focused on preventing outsiders from gaining access to internal company networks, file systems and databases containing sensitive, proprietary data. However, defending organizations from unwitting employee security breaches, compromised accounts, and insider attacks is becoming a growing concern.

Solving the complete IP security problem calls for technologies that protect source data from internal security weaknesses and provide early-warning alerts for risky and anomalous internal behavior. However, security solutions must also take into account the multi-site collaborative nature of today’s design teams.

To successfully protect IP design data from within, companies must look to technologies that support the concepts of both IP/file-level security and data-centric threat detection. This requires two foundational elements:

Firstly, IP-Level and file-level security through an IP management platform such as Methodics ProjectIC that provides IP-level permission assignments and tracking of design data according to IP parent/child relationships, IP branches, levels of hierarchy, and the tracking of who is using which data, where in the design, and in which geographic locations. Once the appropriate permissions are set, the IP management platform will pass this information to the underlying data management system such as Perforce Helix, which then assures the data is secured at repository, branch, directory and even the at an individual file level.

Second, big-data-centric threat detection such as provided by Perforce Helix Threat Detection that offers behavioral analytics and identification of threats and risky behavior performed on Helix SCM repositories.

ProjectIC already has a lot of security features at the project level, smoothly merged into the hierarchy and geographical multi-site aspects of most companies. ProjectIC also works with the underlying design data management tool, such as Helix, to ensure file-level security (so that IP level security cannot be bypassed by simply identifying and copying the underlying files).

The second critical element in a robust IP security solution is the ability to detect anomalous behavior and threats. This is where Perforce Helix Threat Detection comes in, offering a new approach to threat detection. Helix applies advanced big data behavioral analytics to user activity to detect potential attack events, alert security teams, and quickly generate actionable reports that detail anomalous, high-risk behavior.

User activity log files are ingested by the Helix Threat Detection Engine that correlates and analyzes: login and logout, project and file access (folder, file, path, etc), amount of data moved or synchronized (get, commit, sync, etc), timestamp and user data. It applies analysis models (e.g., activity, statistical and clustering mathematics) to log data. Once a threat is identified, a non-intrusive threat detection agent (endpoint sensor) can also be deployed to a laptop or desktop to capture all activity on the endpoint: file copies, cut and pastes, screen captures, printing, obfuscations and exfiltration.

The type of threats that this identifies include:

  • Compromised, careless, and departing employees who download large amounts of data from sensitive projects
  • Insiders who slowly take small amounts of data over a long period of time
  • Machines compromised by stealth malware that are siphoning data
  • Outside or Advanced Persistent cyber attacks

This approach gives companies a double-walled approach to securing proprietary IP data, at both the IP/file level and also head off potential security breaches of theft, whether the threat is internal or external.

The Methodics white paper Threat Detection: A Proactive Approach to Securing SoC IP Design Data is here.


SoC and Foundry Update 2H 2015!

SoC and Foundry Update 2H 2015!
by Daniel Nenni on 09-01-2015 at 10:00 pm

Rarely do I fly first class but I did on my recent trip to Asia. It was one of the new planes with pod-like seats that transforms into a bed. The flight left SFO at 1 A.M. so I fell asleep almost immediately missing the first gourmet meal. About half way through the flight I found myself barely awake staring straight up and what do I see? STARS! That has got to be one of the last things anyone wants to see while looking up on an international flight! Seriously, who puts fake stars on the ceiling of an airplane! EVA Airlines that’s who!

When I travel a lot of people want to meet with me to get the latest news from Silicon Valley. In exchange I get the latest news from wherever they are so it is a very nice quid pro quo type of thing, absolutely. The most common topics are the SoC and foundry business since they currently drive the semiconductor industry. Apple and Qualcomm are the most talked about SoC companies but Mediatek, Samsung, and even Intel are always discussed.

Let’s start with Apple: The big iProduct announcement is next week and we will finally get to see what is inside the iPhone 6s! Again, my bet is a Samsung based 14nm A9 SoC and inside the new iPads will be a TSMC based 16nm A9x SoC. I was right on the iPhone 5s (Samsung 28nm) and iPhone 6 (TSMC 20nm) so let’s see if I can keep my streak going. My bet is also that the Apple A9x will outperform all other SoCs and will continue to do so until mid to late next year.

Moving forward it is my bet that Apple will continue with TSMC 16nm for the iPhone7 with an enhanced version of the process specifically for Apple. Based on what I know today 10nm will not be in production in time for the iPhone 7 but could make it for the next iPads since iPads come out later in the year and require less volume. Currently Samsung and TSMC both have pre-production 10nm PDKs available but final decisions by the fabless elite have not been made. We should know more about where the fabless elite will fab 10nm at the end of this year. I would not expect 10nm production to start before Q2 2017 as there have been delays. The iProduct refresh in 2017 however will be 10nm for sure.

QCOM has a history of 2[SUP]nd[/SUP], 3[SUP]rd[/SUP], and even 4[SUP]th[/SUP] sourcing chip manufacturing down to 40nm. At 28nm everyone was forced into a monogamous relationship with TSMC which was very uncomfortable for a promiscuous company like QCOM. At 28nm QCOM is now in production at UMC and hopes to get ramped up at SMIC to appease the Chinese gods. QCOM as we have all heard will use both Samsung 14nm and GlobalFoundries 14nm for the next generation of Snapdragons. I’m also told that QCOM will use TSMC 16FF+ and they have a 14nm development agreement with SMIC in process.

Mediatek of course manufactures next door (literally) at TSMC and UMC and I do not see that changing anytime soon. Mediatek has hit semiconductor rock star status in Taiwan and they have attracted many ex TSMC and UMC employees. Not only does this give Mediatek leading edge design experience, it also gives them access to the inner foundry ranks. Given the importance of low power design for mobile I would bet Mediatek products will be FinFET enabled next year with the rest of the fabless elite so watch out QCOM!

I’m sorry, I ran out of space for more commentary. If you have questions we can continue the discussion in the comments section. Only registered SemiWiki members can read or write comments so if you are not already a SemiWiki member please join as my guest: https://www.legacy.semiwiki.com/forum/register.php


Adding NAND Flash Can Be Tricky

Adding NAND Flash Can Be Tricky
by Tom Simon on 09-01-2015 at 4:00 pm

As consumers, we take NAND flash memory for granted. It has worked its way into a vast array of products. These include USB drives, SD cards, wearables, IoT devices, tablets, phones and increasingly SSD’s for computer systems. From the outside the magic of flash memory seems quite simple, but we have to remember that this is a technology that relies on quantum tunneling.

For long term data storage, the spinning hard disk has been difficult to beat. I must confess I remember when bubble memory was going to topple flying head hard drives. But that never happened, but now decades later we are seeing a wholesale shift to flash. If you look at Apple’s computer line up, most have flash drives – either hybrid or pure SSD. Their operating speeds are high enough that they can take advantage of PCIe instead of SATA for their interface in some cases.

Now, what are the potential drawbacks of NAND storage devices? The method for writing to flash memory requires passing an electric charge to a floating gate that is situated between the MOS FET gate and the channel. There are two mechanisms for creating this floating charge, quantum tunneling and hot carrier injection. They both involve relatively high voltages, which causes degradation over time of the insulation for the floating gate.

NAND flash when used with direct addressing will have write failures after tens of thousands of cycles. The solution for this is wear leveling, where the physical location of a block of data is moved on every write. This avoids having frequently written blocks, such as OS file directories, wearing out before the blocks that are rarely written to. There are additional enhancements to this to ensure that blocks that are ‘static’ are also moved periodically to use their physical locations as well for necessary write operations, thus spreading around the load.

The relocation and remapping of blocks requires the implementation of fairly complex algorithms. Designers have a choice of using the system CPU for this or offloading the job to a dedicated controller. There are a series of trade offs in the selection of raw NAND, managed by the system CPU, versus a hardware wear leveling and block management solution like what is found in eMMC.

Interestingly, I backed a Kickstarter for the NEEO, a home automation controller, that just posted a blog about a delay in their system design. They had opted for raw NAND in their prototypes but started seeing failures after continuous stress testing. Early on, a potential investor had casually remarked that they ought to look at eMMC. They say in their blog that they owe them a dinner.

Designing embedded controllers for SD and eMMC requires making a number of choices and selection of the proper IP for the protocols that need to be supported. Cadence recently posted a video on their White Board Wednesday video series that give an overview of the technology available to designers from their IP portfolio. Lou Ternullo, Product Marketing Director in Cadence’s IP Division outlines the various protocols and flash technologies they support.

If you are interested in other areas where Cadence offers IP – and there are quite a few – I suggest looking at their IP Factory Brochure. Also their web site features their IP offerings here.

As for flash memory, it is safe to say that its use will continue to expand. Some day in the future we’ll look back at the idea of keeping our valuable data spinning on mechanical merry-go-rounds at 5,000RPM as quaint and primitive.


TSMC is the Top Dog in Pure-Play Foundry Business

TSMC is the Top Dog in Pure-Play Foundry Business
by Pawan Fangaria on 09-01-2015 at 12:00 pm

We all have echoed the fact that the arrival of fabless business model in the semiconductor industry has transformed it completely. The book, “Fabless: The Transformation of the Semiconductor Industry” provides several stories around that. In the backdrop of that, one key point to ponder upon is the start of pure-play foundries; TSMC being the initiator in 1987. The availability of pure-play foundries gave the boost and courage to small as well as large players around the world to start designs without owning fabs. The net result was a flood of fabless design companies and innovations in designs around the world. This is not without the pure-play foundries innovating themselves too. TSMC and subsequent foundries provided leading edge processes and technologies in manufacturing. Today, pure-play foundries provide manufacturing services not only to fabless companies but also to IDMs. Hence, looking at the other side of the coin, it would not be imprecise to say that the pure-play foundry model also transformed the semiconductor industry.

After TSMC, about ten more pure-play foundries were founded around the world, the latest being GLOBALFOUNDRIESin 2009. According to the first quarter sales figures of 2015, three pure-play foundries (TSMC, GLOBALFOUNDRIES, and UMC) occupy the ranks within top20 semiconductor companies in the world, TSMC being 3[SUP]rd[/SUP]. Interesting to note among the pure-play foundries is the following –

The percentage share of these top3 pure-play foundries clubbed together in the overall pure-play foundry sales is 79% and above; $33.68 B out of a total of $42.4 B in 2014, and $9.32 B out of a total of $11.4 B in 1Q 2015, i.e. 82%.

If we do a further analysis of TSMC’s share among the top20 pure-play foundries (i.e. the three foundries as stated above), it’s 74% in 2014 ($24.976 B out of $33.68 B) and 75% in 1Q 2015 ($6.995 B out of $9.318 B). What do we call TSMC in such a scenario?

Let’s also see the pure-play foundry business in recent perspective where we know TSMC had lost some business due to Samsungstarting in-house manufacturing of its Exynos and Appleallocating a part of their processors to Intel and Samsung. However, Apple is expected to come back to TSMC’s 16nm FinFET for their A9 processors in iPhone7. There are reasons for it; I’m not going in those details here. However, I would like to debate on how TSMC influences the overall pure-play foundry business. Let’s look at the following chart reported by IC Insights


This chart depicts the usual trend of the best growth for pure-play foundries in Q2 every year (double-digit growth compared to Q1), i.e. ahead of Q3, the best quarter for total IC industry. However, in 2015 that trend was broken; in Q2 the sales declined slightly compared to Q1 instead of increasing as was seen in previous years. The reason – 5% decline in TSMC’s revenue in Q2 compared to Q1. If TSMC’s 5% change in revenue can change the pure-play industry trend, then that’s definitely the ‘Top Dog’ in the industry. Although there are competing technologies possessed by other foundries as well, I would go back to my hypothesis that business leadership along with technology leadership is the key to establish someone as the ‘Top Dog’.

For TSMC, rest of this year and 2016 are certainly looking better. IC Insights forecasts the overall pure-play foundry sales in Q4 2015 to reach over $12 B, the highest ever. The IC Insights pure-play foundry report is HERE for your reference.

Also read:
Changing Trends at the Top of Semicon Space. The chart in this article provides the sales numbers of the top3 pure-play foundries mentioned above.

Pawan Kumar Fangaria
Founder & President at www.fangarias.com


Solido Wrote the Book on Variation

Solido Wrote the Book on Variation
by Paul McLellan on 09-01-2015 at 7:00 am

When I studied mathematical analysis, one of the things that we had to prove turns out to be surprisingly difficult. If you have a continuous function and at one point it is below a line (say zero) and at another point it is above zero, then there must be a point at which the value is exactly zero. In effect, a continuous function can’t get from below a line to above a line without crossing the line. OK, mathematicians like to spend time proving things that are “obvious” since sometimes they turn out not to be.

How about this, more relevant to semiconductor design. If you simulate a design at the SS corner and the FF corner for some particular parameter, then any other corner will fall between those two values. I mean to get from slow to fast you have to go through the other corner right? Isn’t it obvious? Wrong.

Variation causes weird things to happen. It was not a problem at 90nm but from 28nm on downwards you can’t just simulate those big FF and SS corners and get away with it. Those simulations (at a given voltage and temperature) will define some sort of range but you can’t go from there to the assumption that any other corner will fall inside this range. It is as if you can get from one side of the line at SS to the other FF, without going through typical.

For example, above are a few hundred simulations of a PLL duty-cycle at all sorts of corners including SS and FF. So all the other values “should” fall in between. But look at the distribution. SS is the dot at the far left, so that is pretty much where you would expect to find it. But FF is in the middle of the distribution. If you made the assumption that all other process corners would fall between those two points you would be very wrong.


So it is clear that if you are designing complex analog designs in 28nm or below then you need to do all those simulations to find out what the real distribution is. In the diagram on the left is a simply non-variation-aware flow. On the right is a flow starting to take variation into account. Just pick all the PVT corners that you need and do the simulations. The trouble is that this is prohibitively expensive. In certain cases, such as memories, where these problems are at their worst (there are bit-cells, rows, columns, sense-amps, and more) then the only way to be sure if all you use is brute force is to do a billion simulations. In simpler cases it might be thousands. Words like geological time scale and age of the universe spring to mind. That is not going to be the way to handle this problem.


What is required is a better way to manage this process so that only a subset of the simulations are done. The flow becomes closer to pick some good corners, do the simulations and then see what has been learned. Pick some more corners. Continue until confident that all the important corners have been simulated. The problem is that this cannot be done by hand, it requires a tool to manage the process and do the machine-learning. The diagram above shows a little detail. On the left is the old manual process of simulating a predetermined list of corners. On the right we add intelligence and analysis.

All these diagrams come from the book Variation-Aware Design of Custom Integrated Circuits: A Hands-on Field Guide by Trent McConaghy, Kristopher Breen, Jeffrey Dyck and Amit Gupta of Solido. I should emphasize a couple of things about it. This is not some theoretical analysis of variation for research groups, it is a practical guide for actual design groups. And it is not a user guide to Solido’s tools, it is a guide to what needs to get done, in some sense what needs to get simulated, and while I’m sure Solido are not going to complain if you decide to use their tools, the book is useful even if you do not. It treads that balance between being deep on theory (and thus of little use to a practical designer) and being an extended application note on Variation Designer (and thus of little use to anyone who is not a hands-on user).

The book is available on Amazon here. There you can also get a free sample and you can even try it free (on any Kindle including phones and tablets) for a week.

For anyone who is interested (get a life!) then the proof of the continuous function problem I started with relies on another more primitive fact: any bounded set of real numbers, perhaps infinite, has a least upper bound. Once you have that, then the continuous function problem is easy. The set of values of x that result in a value less than zero must have a least upper bound. But the limit of the function as this bound is approached is zero (that is the definition of a continuous function in math-speak). So there is a value at which the function is zero. W[SUP]5[/SUP]


Xtensa core in Qualcomm low-power Wi-Fi

Xtensa core in Qualcomm low-power Wi-Fi
by Don Dingee on 08-31-2015 at 4:00 pm

Wi-Fi has this reputation as being a power hog. It takes a relatively big processor to run at full throughput. It is always transmitting all over the place, and it isn’t very efficient at doing it. Most of those preconceived notions arose from older chips targeting the primary use case for Wi-Fi in enterprise and residential environments. There is a wall-powered, always-on access point for high performance clients like PCs and smartphones to connect with, streaming data faster and faster.

Those Wi-Fi SoCs were optimized for performance, not power. The pervasiveness and ease of IP connectivity drove many companies to try to reduce Wi-Fi power and footprint to make it more compatible with embedded devices, and more recently, maker modules and IoT devices. Some have been more successful than others – particularly Atmel, Electric Imp, GainSpan, Microchip (from the ZeroG acquisition), Redpine Signals, Silicon Labs, and TI come to mind.

There is a new specification in development called IEEE 802.11ah. It changes frequencies from the 2.4 or 5GHz bands into the sub-GHz band, giving it better range and an improved link budget. Several power-saving mechanisms are being considered, such as Target Wake Time which allows napping until a transmit slot arrives. Proponents of other wireless sensor network technology rained all over the idea, saying competing specifications already deliver the same benefits without all the hassle.

However, none of the companies we’ve mentioned so far are Qualcomm.

One of the more fascinating aspects of our research for Chapter 9 of our upcoming SemiWiki book on mobile has been just how much resistance Qualcomm got from industry naysayers in the early days of CDMA. It was said to “violate the laws of physics”, and it could never be shrunk into a mobile handset, and it wouldn’t work reliably under real-world signaling and loading. Fortunately, there were a few supporters like William C. Y. Lee who understood how powerful digital encoding technology and the Viterbi decoder were. We see how that turned out, though it did take a few years and a detour to another processor after the first six million chips.

Klein Gilhousen of Qualcomm said a few years back, “We have always recognized that the key to feasibility of CDMA was VERY aggressive custom chip design.” We have been saying for some time here at SemiWiki that we are waiting for SoCs specifically designed for IoT tasks to appear. If there is a company that can pull off low-power Wi-Fi designs, it’s Qualcomm – just substitute “IoE” (the term Qualcomm prefers) for “CDMA” in Gilhousen’s statement.

We are seeing what the Atheros acquisition does for Qualcomm. In a world where mobile growth rates are settling down and IoT is the next field of opportunity, Qualcomm has chips for low-power Wi-Fi. Note these designs are not for 802.11ah as of yet – Qualcomm is a sponsor of the spec, but it is still in early development.

Rather, Qualcomm has taken experience from the CDMA portfolio and added new ideas for Wi-Fi in existing 2.4/5GHz settings, including its AllJoyn thin client software. For example, let’s look at the QCA4004, implementing 802.11n 1×1 SB/DB low power Wi-Fi. It is flexible enough for three Wi-Fi use cases:

The first is the traditional one, a peripheral to a host. The second is the embedded case, with a local MCU. The third is the IoE case where the onboard core is powerful enough to run AllJoyn.

Atheros first licensed the Tensilica Xtensa core back in 2005, way before Cadence acquired Tensilica in 2013. The QCA4004 was introduced almost two years ago, and its press release and datasheet have no details on the processor core inside. A more recent Qualcomm presentation at IoE Day 2014 in Shenzhen reveals a Cadence Xtensa core at the center of the QCA4004, running at 130MHz with the ThreadX RTOS from Express Logic. That allows the same software architecture to be used whether an external MCU is present or not, including a dual IPv4/IPv6 stack (take that, TI) and AllJoyn API.

The Qualcomm influence is more evident in two areas of the QCA4004 design. The first is the antenna diversity feature, with an RSSI algorithm in hardwired DSP logic to determine which antenna gives the best results. The second is a Green Tx feature, allowing the Tx power to be reduced to the lowest acceptable level for a given throughput. The active power figures for the QCA4004 are worth noting. At 2.4GHz, Tx power at 18dBm is 230mA, while Rx power is 60mA. Sleep mode gets down to 130uA.

The core choice of Cadence Xtensa puts Qualcomm in new territory for smaller, cheaper, more configurable chips for low-power Wi-Fi. Naysayers beware. While other solutions might have traction in industrial IoT environments, the combination of Wi-Fi ubiquity, the right chip design expertise, AllJoyn software, and an ecosystem that includes Microsoft makes Qualcomm a strong player in consumer IoT use.

We’re tracking what else Qualcomm is up to as they remake the company.


Optimizing SRAM IP for Yield and Reliability

Optimizing SRAM IP for Yield and Reliability
by Daniel Payne on 08-31-2015 at 12:00 pm

My IC design career started out with DRAM at Intel, and included SRAM embedded in GPUs, so I recall some common questions that face memory IP designers even today, like:

  • Does reading a bit flip the stored data?
  • Can I write both 0 and 1 into every cell?
  • Will read access times be met?
  • While lowering the supply voltage does the cell data retain?
  • How does my memory perform across variation?

If you are buying your SRAM IP then maybe you don’t have to be so concerned about these questions, however the circuit designers responsible for design and verification of memory IP are very focused on getting the answers to these. Consider the typical, six-transistor SRAM bit cell in a CMOS technology:

The basic memory element has a pair of cross-coupled inverters (devices PD1, PL1 and PD2, PL2) to store a 1 or 0, then to read or write the cell you activate the Word Line (WL) through NMOS devices PG1 and PG2. A circuit designer chooses device sizes and can start to optimize this SRAM bit cell by running a worst-case analysis where one objective is to minimize the read current path shown above in the red arrow. During this worst-case analysis the Vth for devices PG1 and PD1 are adjusted iteratively, your favorite SPICE circuit simulator is run, then the results can be determined by a tool like WiCkeDfrom MunEDA:

The axis of this plot are the variations in Vth for devices PG1 and PD1, while the circles represent the amount of variation measured in sigma. The read current contour values are shown as dashed lines. To minimize the read current we look at the intersection of the dashed lines and circles, so there’s a Red dot showing our worst-case point within a 6 sigma distance.

Related – High Sigma Yield Analysis and Optimization at DAC

Moving up one level of hierarchy from a single memory cell to the actual memory array we have a typical architecture that combines cells into columns, where at the bottom is a mux, equalizer and Sense Amp (SA) circuit:

To run a hierarchical analysis and understand how variation effects this SRAM there are challenges:

[LIST=1]

  • For each SA there are N cells
  • In Monte-Carlo sampling we have to take N cell samples for each SA sample
  • Count the failure rate whenever t[SUB]SA[/SUB] > t[SUB]max[/SUB]

    Brute Force Sampling
    One approach is brute force sampling, so if our SRAM had 256 cells per SA and we wanted 10,000 results, that would require 256 * 10,000 simulation runs in a SPICE circuit simulator. Using a brute force Monte-Carlo analysis for SRAM design isn’t really feasible for anything beyond 4 sigma, because it would require millions to billions of SPICE runs, something that we don’t have enough time to wait for.

    Scaled Sampling Approximation
    Another method is for the circuit designer to use only 4 sigma variation in the SA while the memory cell has 6 sigma. This approach takes less effort than brute force and is easier to run, however it provides an incorrect approximation.

    Two Stage Worst-Case Analysis (WCA)
    The recommended approach by MunEDA is to first calculate the 6 sigma worst-case condition of the cell using the Voltage on the Bit Line (VBL) at the moment that the SA is enabled. The second stage is to calculate the 4 sigma worst-case condition of t[SUB]SA[/SUB] for the sense amp, equalizer and MUX. Here are two charts showing the SA offset versus cell current for a variation where the worst-case point is in spec (green region), and out of spec (red region).


    Worst-case point is in spec, then out of spec

    You can also compare a sampling approach against the two stage WCA by looking at the following charts for SA offset versus cell current:


    With the sampling approach it estimates the failure rate by using sampling points, Red dots for failing and Green dots for passing. On the downside sampling relies on tail distribution accuracy, and suffers from sampling error. The distribution of local variation variables in the tails with >5 sigma is not well characterized, so the true tail distribution in silicon can differ significantly from the ideal Gaussian distribution used inside the model files. Running a global process monte-carlo is not a guarantee to cover the full corner range that can be seen in silicon.

    So a large local plus global monte-carlo run is infeasible because of long run times, plus it’s sensitive to distribution errors. Even speeding up monte-carlo is not really sufficient because it will just produce unsafe results in a shorter period of time. So, what we really need is a new method that can:

    [LIST=1]

  • Handle the large, structured, hierarchical netlist of SRAM arrays.
  • Adjust conservatism in the local variation tails
  • Run quickly, so that local variation analysis can be repeated over multiple PVT corners, design options and layout options


    With two stage WCA we are estimating the failure rate by a large sample Monte-Carlo approximation in the pink region, using a conservative estimate in the pink plus green region, and showing the worst-case point check as passing by a Green dot. The tool flow GUI in WiCkeD makes it quite easy for a circuit designer to specify their own memory array size, target failure rate, and to trade off the array failure rate with read time:

    Comparing all three analysis techniques for a 28nm SRAM block show how the two stage WCA approach uses the least CPU effort in SPICE circuit simulations, scales well to high sigma analysis, and has results close to full Monte-Carlo:

    Related – When it comes to High-Sigma verification, go for insight, accuracy and performance

    Summary

    It is possible to design, analyze and optimize SRAM IP blocks using a two stage WCA approach, while taking much less circuit simulation time than a brute force Monte-Carlo, and at comparable accuracy. All you need to add into your existing EDA tool flow is the WCA capabilities in the MunEDA WiCkeD tool.

    To find out more about MunEDA, there’s a 30 minute webinar coming up on September 9th at 10AM (PST), register here.


  • Michael Sanie Plays the Synopsys Verification Variations

    Michael Sanie Plays the Synopsys Verification Variations
    by Paul McLellan on 08-31-2015 at 7:00 am

    I met Michael Sanie last week. He is in charge of verification marketing at Synopsys. I know him well since he worked for me at both VLSI Technology and Cadence. In fact his first job out of college was to take over support of VLSIextract (our circuit extractor), which I had written. But we are getting ahead.

    Michael was born in Iran and came to the US as a teen. He was a very good pianist, good enough to go on music tours and appear on TV shows, an experience he describes as “not necessarily recommended for any kid”. He expected to go to a music conservatory but instead he went to Purdue and studied CEE (what everyone calls CS/EE today). As he likes to say “I studied to be a conductor, and I must have been only semi-good, as I ended up in semi-conductors.”

    While supporting VLSIextract and more, Michael went to Santa Clara and got an MBA and also made the transition from engineering into marketing.

    In 1999 he joined Numerical Technologies, one of the many DFM startup companies of that era, doing marketing and business development. It was the most successful of those DFM startups, going public in 2000 and being acquired in 2003 by Synopsys for $250M.

    Next stop was Cadence where he did various things, including as I mentioned working for me for at time in strategic marketing.

    After a couple of years at Cadence, Michael moved to Calypto to be their first VP marketing. They were bringing to market sequential formal verification technology, which was new and innovative technology. Michael drove the first deals and also created a partnership with Mentor which eventually ended up with Mentor taking a majority position in the company and adding their Catapult synthesis to the product portfolio. Of course Calypto still exists today.

    Michael joined Synopsys almost six years ago and joined the Verification business group, which has grown steadily to now offering simulation, emulation, static analysis, formal, debug and more.

    Michael reckons that verification is an area where lots of innovation is happening, at least partially because you can never have enough verification. There has been a big shift in the last few years towards emulation, which has gone from an esoteric segment for the occasional group who can afford a seven-figure sum to part of the mainstream. It has also turned out to be the complementary piece to virtual prototyping.

    A lot of the focus of verification these days is to make sure software can be run before tapeout. This is firstly so that the software engineers can start development before silicon is back from the fab. But it is also because if your chip has to run, say, Android, then you really really want to boot Android on a model of the chip before tapeout, no matter how much verification has already been done. Though 90%+ of bugs are still found by simulation, booting Android takes around 40B cycles so you are going to need emulation for it. .

    The next step looks like systematic power analysis in the context of verification, in particular emulation with power. So not just “can I boot Android?” but “can I boot Android without frying the chip?”

    Another challenge is how best to intelligently combine formal and simulation. If formal verification has proved something, then it is not necessary to prove it again with simulation. On the other hand, simulation results can be intelligently used to lead to good assertions for use in formal.

    Synopsys recently expanded its verification investment by acquiring Atrenta, and now they are looking at how to combine static analysis with other techniques to get results faster. One great things about Spyglass (Atrenta’s anchor platform) is that it is used on RTL when the design has only just started to be developed, long before it is in any sense complete. Early feedback on power, DFT and physical issues while the RTL is being created.

    Verification does have a big and direct impact on time-to-market, especially due to software bringup. In the days when you had to wait for the chip before software development started, versus today, the schedule can be pulled in by weeks if not months.

    On emulation, as you probably know, Synopsys acquired EVE a couple of years ago. The ZeBu hardware technology was and is based on Xilinx FPGAs so that they can leverage Xilinx’s R&D investment with faster chips every 2-2½ years, and with lower cost than designing their own chips. Many industry leaders are aligning with this strategy as it enables them to keep up with their demands for performance, capacity, and cost. Historically one challenge with emulation (any emulator) is that it took a good amount time and effort (including possible changes to the RTL) to bring a design up in emulation. Synopsys has unified the VCS compile process with ZeBu to significantly shorten this bring up time. Also as another result coming from this project, ZeBu compile time is 3x faster (most designs only take 4 hours and often much less) and if it compile in VCS is should compile on the ZeBu. Synopsys are still #3 in emulation but bridging the gap fast.

    VIP is another area that is doing well since the interfaces have got so complex that no design group can do verification on their own. For example, the latest USB 3 spec is insanely complex, and there are a dozen interfaces like that. There is an obvious synergy with Synopsys’s IP business since they are #1 in interface IP.

    Of course Michael is not going to make some product announcement in the middle of an interview like this, but he did say that, as you would expect, Synopsys is working on all of these areas. “Watch this space” as the saying goes.

    Yes, Michael continues to study piano although with his travel schedule he can’t exactly go on concert tours., He studied for many year (started when he was at VLSI) with a former professor of the Vienna Conservatory. More recently, he’s been taking lessons at Stanford. It turns out that there are a surprising number of people at Stanford studying both music and engineering. Apparently math, music, language and computer science all use the same part of the brain so often people who are good at one on the list are good at another.

    See also John Koeter: How To Be #1 in Interface IP
    See also Synopsys’ Andreas Kuehlmann on Software Development
    See also Antun Domic, on Synopsys’ Secret Sauce in Design
    See also Bijan Kiani Talks Synopsys Custom Layout and More


    China: drag on global semiconductor market?

    China: drag on global semiconductor market?
    by Bill Jewell on 08-30-2015 at 7:00 pm

    The Chinese stock market (as measured by the Hang Seng Index) dropped 11% from August 14 to August 24 over concerns of a slowing economy. In reaction, the U.S. stock market (as measured by the S&P 500) dropped 11% from August 17 to August 25. The China market has since rebounded 2% while the U.S. market rebounded 5%. Will a slowing China drag down the global economy? China accounts for about half of the global semiconductor market. Will slowing semiconductor demand in China lead to a major slowdown in the global semiconductor market?

    Chinese electronics production data presents a mixed picture as shown below. Unit production three-month-average change versus a year ago shows mobile phones went negative in March 2015. Mobile phones also went negative in 2012 before bouncing back to strong growth in 2013. PCs turned negative in September 2014, reflective of the weak global demand for PCs. Televisions were negative in December 2014 and May 2015, but have generally had modest growth in 2015. In contrast to the volatility of unit production, the overall change in electronics production (communications equipment, computers and other electronics) as measured in Chinese yuan has been steadier. Overall electronics production growth has been below 10% for the last three months after averaging 12% for the years 2012 to 2014.


    Over the last ten years, China GDP growth has been gradually slowing down. Following double-digit growth in 2006-2007, GDP dropped to 9.6% in 2008 and 9.2% in 2009 – still strong growth especially since most of the rest of the world was experiencing a major recession. Growth picked back up to 10.4% in 2010 and moderated to 7.4% in 2014. The International Monetary Fund (IMF) projects China GDP will continue to decelerate to 6.0% in 2017 before increasing to 6.3% in 2019 and 2020. China electronics production growth (as measure in yuan) has averaged 4 percentage points above GDP growth from 2006-2014. Our forecast at Semiconductor Intelligence (SC IQ) is electronics will grow in the 9% to 10% range through 2020.


    Thus China should still be a major growth driver of the global economy and semiconductor market for at least the next few years. The behavior of stock markets is almost impossible to predict and difficult to explain. Stock markets are influenced by economic factors, short term computer trading, greed and fear. We believe the recent behavior of the China and U.S. stock markets are not a sign of a significant slowing of the growth of the Chinese economy in the next few years.