Duncan Bremner, Founder and CTO, SureCore Limited

Thanks to the film ‘Mean Girls’,  we can all remember how it was at school; the geeks were geeking, the sports jocks were jockeying, the plastics were strutting their stuff, and everyone else was just trying to get through the  day. Then BAM!!;  a new, in your face, upstart appears on the scene, ruffles everybody’s feathers and challenges the status quo.

The same has just happened to the technology markets with the arrival of the Internet of Things (IoT). No-one really knows what it is, how it will change their business models, and what opportunities it may uncover. At present, few have even tried to understand the global implications[1]. The one thing everyone is convinced of is the arrival of IoT will be a significant market inflection point; Cisco has suggested it will have an impact 10 times greater that the internet itself[2].

The semiconductor industry is well positioned to benefit from the arrival of the IoT but only if it embraces and adapts itself to the changing demands that this emerging market will place upon it. In particular, the IoT will place much greater emphasis on power efficiency, both in high performance, data mining applications within servers and networking clusters as well as in the end user, power sensitive and sensors nodes.  Here we examine the challenges to the semiconductor industry of delivering components and sub-blocks optimized for the mass adoption of power critical devices while delivering considerably greater computing and storage resources into previously, relatively dumb devices.

surecore 1

One of the most demanding areas for low power design and manufacture is embedded SRAM. According to Semico Research, embedded SRAM now consumes up to two-thirds of the area of new System–on-chip (SoC) devices and will exceed 70 percent by 2017. As the market for IoT end nodes develops, the integration of very low power storage will be critical to the success. Many applications must rely on renewable or energy harvested energy sources, demanding the re-design of memory sub-blocks to improve both dynamic and static (leakage) consumption.  This will lead to process bifurcation; low power applications such as power sensitive end nodes will stay on 65/40nm processes while high performance applications such as data and network traffic processing will drive technology. Today’s deep sub-micron processes are already being challenged on power as evidenced by TSMC’s announcement of their Ultra-Low Power (ULP) processes on 55, 40 and 28nm.  As the industry continues towards 16nm technology, this split will become more pronounced as demand for greater functionality and extended battery life grows.

IoT applications will demand more for less on two fronts: more battery life (less power consumption) combined with more local storage capability while maintaining or reducing the end product price point. When further constrained by the demand for environmental friendly technology, this challenge reaches crisis proportions, affecting both low power and high performance application market segments. Several of these emerging applications, such as remote and distributed sensing applications for buildings, will be supported by the replacement market within the construction industry.   However, some of the emerging trends for the Internet of Things (IoT) such as Wearables, home health and other novel markets are an anathema to semiconductor marketers.

Over the last two decades, embedded SRAM has followed Moore’s law delivering area, power and performance benefits. However, at 40nm and below, power density and power management considerations have become increasingly important and the industry’s response to meet these demands was to reduce the operating voltages below 1V.  When combined with atomistic variability[3], this required additional circuitry to perform “read assist” and “write assist” to overcome the intrinsic bit cell stability issues.

These challenges encouraged the engineering team at sureCore Limited to examine the fundamentals of SRAM architecture and, by developing novel circuit design and architecture techniques, they have addressed static and dynamic power consumption while also improving variability tolerance. These techniques have resulted in market leading power savings. This highly power efficient SRAM memory IP now maintains its data resilience while operating at low supply rails irrespective of the process technology and thus is ideally suited for IoT application requirements.

While the impact of power management will be felt throughout the industry, the two pathfinder applications identified by sureCore occupy the two power management extremes.

The first covers new devices addressing IoT remote sensing applications. According to Gartner[4], the demand for IoT silicon in endpoint devices will grow 36.2 percent in 2015 against an industry growth figure of 5.7 percent. They go on to say that, when taken in context, this sector will drive the overall semiconductor market. This use case is characterised by small physical size, low cost and the ability (ideally) to operate without batteries; energy instead being supplied from renewable sources such as light, heat, or mechanical energy and the circuit’s power needs ‘scavenged’ or ‘harvested’ from these sources.  As many of the IoT applications are designed around small, self-contained, wireless connected, low-cost sensors nodes, the power challenge becomes a system level problem; trading the ability to store and process information locally against the power demands to enable wireless communication to a more computationally intelligent, centralized node. Based on these requirements, Gartner suggests that processing will be the largest revenue contributor with $7.58 billion while sensors will be the fastest growing with 47.5 percent growth, both in 2015.

The second use case is in high performance, computationally-intense applications such as graphics processors, network processors or network routing engines.  In these applications, the challenge stems from on-chip power dissipation generating excess heat that must be transferred away from the silicon component to increase reliability and operational efficiency.  The impact on reliability is well documented but the commercial impact is less well understood and can be split into three separate benefits; operational savings, capital savings and environmental savings.

Firstly, the operational savings realized by the direct equipment energy reduction plus the energy reduction through heat management.  Although the second term in the operational costs varies greatly, depending upon equipment scale and complexity, a reasonable heat management solution for, say, a rack mount server / network processor is conservatively estimated between 50-60 percent of the heat energy generated.  That is, for every 10 Watts of heat generated, it takes around five to six Watts to manage the heat away. Through addressing the fundamental energy consumption through the development and adoption of lower power solutions, this operational expense can be slashed.

Then there’s the matter of capital savings.  Effective heat management requires equipment space, material and real estate.  Reducing power consumption means smaller heat sinks, fans, power supplies and chassis components.  When scaled to a typical server farm, this translates into higher density (or smaller) racks culminating in a reduction of floor area for a given performance metric.  The drive towards cloud computing and storage has resulted in major compute farm operators (Google, Apple etc.) locating their new facilities where low cost electrical power and environmental conditions are favorable.  Iceland, for example, has been identified as an attractive location given its cheap electrical power and low ambient temperature.

The third benefit is environmental.  According to Jeff Monroe, head of Verne Global, a data center company in Iceland, “The data center industry now is on par with the airline industry as far as the carbon footprint.”  As environmental concerns and the increasing costs of energy take hold, organizations will seek to offset their energy consumption through the adoption of energy reduction technologies such as those invented by sureCore; the so-called Green Computing initiatives.

So the question becomes how to accrue these cost savings?  sureCore has conducted a root and branch analysis of power consumption in next generation SoCs.  Looking at the SRAM as a source for power savings, the company has designed low-power SRAM IP that can achieve over 50 percent power savings at comparable performance when compared against industry standard solution.   These savings are available to SoC partner developers today ensuring any IoT products developed using the new low power technology will be in the vanguard of the predicted IoT growth in 2015.

It is imperative that the industry take a pro-active approach to the power requirements of IoT. Ever since the invention of the integrated circuit, the product capabilities demanded by the market have always exceeded the technical ability of the industry. The IoT is the first major market where the industry can deliver the feature set demanded by the market.  Power efficiency is the new mantra adopted by users and it is the responsibility of the semiconductor industry to take heed.  This extends to adopting a holistic view of the application needs to optimize across the whole delivery platform. It is recognized that the industry is at a process crossroads as geometries fall below 20nm. It is unlikely that the same process (albeit with some minor modifications), whether Bulk, FD-SOI or FinFET, will be able to deliver benefits to everyone in the ecosystem. The most likely scenario will be that each of these technology approaches will become optimized to deliver technology and commercial value across different parts of the IoT value chain. However, the new constant driver will be power management and, for memory intensive applications irrespective of technology, sureCore will have a solution to the memory power challenges customers face.

[1] See Knowledge Transfer Network, March 2015; ‘The IoT Tree of Life’
[2] See Cloud Times; 7 March 2014; ‘Cisco Says Internet of Things will Have Ten Times More Impact on Society Than Internet’
[3] Gold Standard Simulations Limited: Atomistic variability simulator
[4] Gartner Inc., Gartner’s 2015 Predictions Special Report