Part of the  

Chip Design Magazine


About  |  Contact

Posts Tagged ‘machine learning’

Memory Subsystem Solutions Announced

Tuesday, May 9th, 2017

Gabe Moretti, Senior Editor

Integrating a Network on Chip (NoC) with a memory controller provides increased system throughput by decreasing the latency of data transfer to and from local storage.  Sonics, Inc. and Northwest Logic have announced their partnership to deliver high throughput memory subsystem solutions for complex System-On-Chip (SOC) designs.  The subsystem is focused on SOC Designs for Machine Learning, Computer Vision, UHD Video Processing, and Enterprise SSD Applications.   The companies’ partnership, which is being driven by a mutual customer SOC design win, integrates Sonics’ flagship interconnect fabric, SonicsGN NoC, and Sonics’ MemMax memory scheduler with Northwest Logic’s family of HBM2, DDRx, LPDDRx memory controllers.

“As DRAM data rates increase, the number of pipelined outstanding transactions required to achieve full throughput grows dramatically,” said Drew Wingard, CTO of Sonics. “This requires more intelligent transaction scheduling that considers both the bandwidth and latency requirements of pending requests and the page and bank states of DRAM. Without careful coordination between the NoC, memory scheduler, and memory controller, the subsystem will miss data transfer deadlines and suffer performance degradation at several points along the memory subsystem transaction path. Our partnership with Northwest Logic ensures that MemMax’s scheduling decisions produce a transaction stream enabling Northwest Logic’s controllers to efficiently map into memory commands that fully leverage the customer’s chosen DRAM technology.”

“Northwest Logic and Sonics share an uncompromising commitment to customer success,” said Brian Daellenbach, President of Northwest Logic. “We are seeing a significant uptick in demand for high throughput memory subsystems that address data-intensive applications and markets such as Machine Learning, Computer Vision, UHD Video Processing, and Enterprise SSD. Our memory controllers have a strong industry reputation for delivering high-performance, high quality, and ease-of-use. Our partnership with Sonics enables us to provide our mutual customers with a complete memory subsystem solution that also takes into account the need for high performance NoCs that support multi-channel memory subsystem architectures and integrate all of the cores in the system.”

The Sonics-Northwest Logic high throughput memory subsystem solutions where developed for a mutual customer and are now being integrated in designs by other customers as well.

EDA in the year 2017 – Part 1

Thursday, January 12th, 2017

Gabe Moretti, Senior Editor

The EDA industry performance is dependent on two other major economies: one technological and one financial.  EDA provides the tools and methods that leverage the growth of the semiconductor industry and begins to receive its financial rewards generally a couple of year after the introduction of the new product on the market.  It takes that long for the product to prove itself on the market and achieve general distribution.

David Fried from Coventor addressed the most important topics that may impact the foundry business in 2017.  He made two points.

“Someone is going to commit to Extreme Ultra-Violet (EUV) for specific layers at 7nm, and prove it.  I expect EUV will be used to combine 3-4 masks currently using 193i in a multi-patterning scheme (“cut” levels or Via levels) for simplicity (reduced processing), but won’t actually leverage a pattern-fidelity advantage for improved chip area density.

The real density benefit won’t come until 5nm, when the entire set of 2D design rules can be adjusted for pervasive deployment of EUV.  This initial deployment of EUV will be a “surgical substitution” for cost improvement at very specific levels, but will be crucial for the future of EUV to prove out additional high-volume manufacturing challenges before broader deployment.  I am expecting this year to be the year that the wishy-washy predictions of who will use EUV at which technology for which levels will finally crystallize with proof.

7nm foundry technology is probably going to look mostly evolutionary relative to 10nm and 14nm. But 5nm is where the novel concepts are going to emerge (nanowires, alternate channel materials, tunnel FETs, stacked devices, etc.) and in order for that to happen, someone is going to prove out a product-like scaling of these devices in a real silicon demonstration (not just single-device research).  The year 2017 is when we’ll need to see something like an SRAM array, with real electrical results, to believe that one of these novel device

concepts can be developed in time for a 5nm production schedule.”

Rob Knoth, Product Marketing Director, Digital and Signoff Group at Cadence offered the following observation.  “This past year, major IDM and pure-play foundries began to slow the rate at which new process nodes are planned to be released. This was one of the main drivers for the restless semiconductor-based advances we’ve seen the past 50 years.

Going forward, fabs and equipment makers will continue to push the boundaries of process technology, and the major semiconductor companies will continue to fill those fabs. While it may be slowing, Moore’s Law is not “dead.” However, there will be increased selection about who jumps to the “next node,” and greater emphasis will be placed on the ability of the design engineer and their tools/flows/methods to innovate and deliver value to the product. The importance for an integrated design flow to make a difference in product power/performance/area (PPA) and schedule/cost will increase.

The role that engineering innovation and semiconductors play in making the world a better place doesn’t get a holiday or have an expiration date.

The semiconductor market, in turn, depends on the general state of the world-wide economy.  This is determined mostly by consumer sentiment: when consumers buy, all industries benefit, from industrial to financial.  It does not take much negative inflection in consumers’ demand to diminish the requirement for electronic based products and thus semiconductors parts.  That in turn will have a negative effect on the EDA industry.

While companies that sell multi-years licenses can smooth the impact, new licenses, both multi-year and yearly are more difficult to sell and result in lower revenue.

The electronic industry will evolve to deal with increased complexity of designs.  Complex chips are the only vehicle that can make advance fabrication nodes profitable.  It makes no sense decreasing features’ dimensions and power requirements at the cost of increased noise and leakage just for technology sake.  As unit costs increase, only additional functionality can justify new projects.  Such designs will require new methodology, new versions of existing tools, and new industry organization to improve the use of the development/fabrication chain.

Michael Wishart, CEO of Efabless believes that in 2017 we will begin to see full-fledged community design, driven by the need for customized silicon to serve emerging smart hardware products. ICs will be created by a community of unaffiliated designers on affordable, re-purposed 180nm nodes and incorporate low cost, including open source, processors and on-demand analog IP. An online marketplace to connect demand with the community will be a must.

Design Methods

I asked Lucio Lanza of Lanza techVentures what factors would become important in 2017 regarding EDA.  As usual his answer was short and to the point.  “Cloud, machine learning, security and IoT will become the prevailing opportunities for design automation in 2017. Design technology must progress quickly to meet the needs of these emerging markets, requiring as much as possible from the design automation industry. Design automation needs to willingly and quickly take up the challenge at maximum speed for success. It’s our responsibility, as it’s always been.”

Bob Smith, Executive Director of the ESD alliance thinks that in 2017, the semiconductor design ecosystem will continue evolving from a chip-centric (integration of transistors) focus to a system-centric (integration of functional blocks) worldview. While SoCs and other complex semiconductor devices remain critical building blocks and Moore’s Law a key driver, the emphasis is shifting to system design via the extensive use of IP. New opportunities for automation will open up with the need to rapidly configure and validate system-level design based on extensive use of IP.  Industry organizations like the Electronic System Design Alliance have a mission to work across the entire design ecosystem as the electronic design market makes the transition to system-level design.

Wally Rhines, Chairman and CEO of Mentor Graphics addressed the required changes in design as follows: “EDA is a changing.  Most of its effort in the last two decades in the EDA industry has focused on the automation of integrated circuit design. Virtually all aspects of IC design are now automated with the use of computers.  But system design is in the infancy of an evolution to virtual design automation. While EDA has now given us the ability to do first pass functional integrated circuit designs, we are far from providing the same capability to system designers.

What’s needed is the design of “systems of systems”.  That capability is coming.  And it is sooner than you might think.  Designers of planes, trains and automobiles hunger for virtual simulation of their designs long before they build the physical prototypes for each sub-system.  In the past, this has been impossible.  Models were inadequate.  Simulation was limited to mechanical or thermal analysis.  The world has changed.  During 2017, we will see the adoption of EDA by companies that have never before considered EDA as part of their methodology.”

Frank Schirrmeister, Senior Product Management Group Director, System and Verification Group at Cadence offered the following observation.  “IoT that spans across application domains will further grow, especially in the industrial domain. Dubbed in Gernany as “Industrie 4.0”, industrial applications are probably the strongest IoT driver. Value shifts will accelerate from pure semiconductor value to systemic value in IoT applications. The edge node sensor itself may not contribute to profits greatly, but the systemic value of combining the edge node with a hub accumulating data and sending it through networks to cloud servers in which machine learning and big data analysis happens allows for cross monetization. The value definitely is in the system. Interesting shifts lie ahead in this area from a connectivity perspective. 5G is supposed to broadly hit is in 2020, with early deployments in 2017. There are already discussions going on regarding how the connectivity within the “trifecta” of IoT/Hub/Server is going to change, with more IoT devices bypassing the aggregation at the hub and directly accessing the network. Look for further growth in the area that Cadence calls System Design Enablement, together with some customer names you would have previously not expected to create chips themselves.

Traditionally ecosystems have been centered on processor architectures. Mobile and Server are key examples, with their respective leading architectures holding the lion share of their respective markets. The IoT is mixing this up a little as more processor architectures can play and offer unique advantages, with configurable and extensible architectures. No clear winner is in sight yet, but 2017 will be a key year in the race between IoT processor architectures. Even OpenSource hardware architectures are look like they will be very relevant judging from the recent momentum which eerily reminds me of the early Linux days. It’s definitely one of the most entertaining spaces to watch in 2017 and for years to come. “


Standards have played a key role in EDA.  Without them designers would be locked to one vendor for all of the required tools, and given the number of necessary tools very few EDA companies would be able to offer all that is required to complete, verify, and transfer to manufacturing a design.  Michiel Ligthart, President and COO at Verific, sees two standards, in particular, playing a key role in 2017.  “Watch for quite a bit of activity on the EDA standards front in 2017. First in line is the UVM standard (IEEE 1800.2), approved by the Working Group in December 2016. The IEEE may ratify it as early as February. Another one to watch is the next installment of SystemVerilog, mainly a “clarifications and corrections” release, that will be voted on in early 2017 with an IEEE release just before the end of the year. In the meantime, we are all looking at Accellera’s Portable Stimulus group to see what it will come up with in 2017.”

In regards to the Portable Stimulus activity Adnan Hamid, CEO of Breker Verification Systems goes into more details.  “While it’s been a long time coming, Portable Stimulus is now an important component of many design verification flows and that will increase significantly in 2017. The ability to specify verification intent and behaviors reusable across target platforms, coupled with the flexibility in choosing vendor solutions, is an appealing prospect to a wide range of engineering groups and the appeal is growing. While much of the momentum is rooted in Accellera’s Portable Stimulus Working Group, verification engineers deserve credit for recognizing its value to their productivity and effectiveness. Count on 2017 to be a big year for both its technological evolution and its standardization as it joins the ranks of SystemVerilog, UVM and others.


Given the amount of contributions received, it would be overwhelming to present all of them in one article.  Therefore the remaining topics will be covered in a follow-on article the following week.