Surfing the hyperscale data tsunami with silicon

Posted on 05/16/2018 by

Bob Stear

The world is already awash in digital information, but what might be surprising is how enormous this virtual tsunami of data is becoming. The Cisco Global Cloud Index predicts a three-fold increase in the annual data center traffic to 20.6 ZB by the year 2021. This stunning amount of data is driven in large part by the explosion of connected devices with a recent IHS Report detailing the exponential growth of the Internet of Things to 125 billion devices by the year 2030. That is almost 15 connected devices per person estimated to be on Earth at that time! A recent report by We Are Social on the state of our increasingly connected world indicates:

  • Two-thirds of the world’s 7.6 billion people have mobile phones.
  • More than half those mobile phones are “smart” devices.
  • Over half the world’s population (4 billion) uses the internet.
  • Nearly 3.2 billion of us are active social media users.

The growing number of internet and mobile device users is impressive, but it accounts for only a fraction of the total data increase. That’s because the amount of data each of us produces and consumes is exploding as well, thanks to our insatiable appetites for high quality audio and video content, social media connections, e-commerce and a multitude of other internet-enabled products and services.

Add on top of all that 4TB per day of data traffic for each autonomous vehicle and 10GB per second of data traffic from new jetliners and you can see how we quickly get to that 20.6 ZB annually by 2021.

 

Shifting data center needs

This data storm is being enabled by a massive scalable information handling infrastructure made up of millions of interconnected computer servers, routers, switches and storage equipment, most of which are housed in huge, air-conditioned data center facilities in “geographically safe” locations around the world.

Data centers were originally used by large enterprise companies to manage their own internal data-processing operations. But most companies these days are outsourcing their information technology needs to remotely located data centers, which are more economical, flexible, and dependable than in-house operations.

Today’s biggest data center operators, the “Super 7”, which includes Alibaba, Amazon, Baidu, Facebook, Google, Microsoft, and Tencent, are racing to build a new generation of “hyperscale” data centers. These new data centers are larger, massively scalable, more energy efficient, cost less to operate, and are designed for the latest advanced computing tools with massive arrays of CPUs, GPUs, TPUs, and FPGAs enabling deep learning, machine learning, big data analysis and key artificial intelligence (AI) processes. Significant effort is also being invested by all the major hyperscale players in research and development of new custom AI processors, delivering more efficient inference with techniques like near memory computing.

Synergy Research Group estimates the five top vendors collectively spent more than $50 billion last year on new hyperscale data centers. It also estimates that the approximately 400 hyperscale facilities operating worldwide at the end of 2017 will pass 500 by 2020.

 

Semiconductor innovation and growth opportunities

With that kind of growth, and with new hyperscale facilities costing upwards of $500 million apiece to build and equip, it’s no wonder that data centers have become one of the biggest, fastest growing markets for servers, networking, data storage and semiconductors. This hyperscale datacenter growth is also driving ongoing waves of innovation in semiconductors, demanding the continued benefits of Moore’s Law through innovation in advanced process technology, high speed / high bandwidth IP, and heterogeneous packaging capabilities.

At Samsung Foundry, we’re helping our customers design and build a wide variety of data center-focused chips from network routers and switches to CPUs/GPUs to storage controllers. These increasingly complex hyperscale datacenter semiconductors will benefit from the leading edge high performance / high integration platform solutions that Samsung Foundry provides including our new industry 1st 7-nm LPP (Low Power Plus) extreme ultraviolet (EUV) technology offering, silicon verified high performance IP like 100+Gbps High Speed SerDes, HMB2, and GDDRx PHYs from Samsung and our SAFETM foundry eco-system program partners, along with an array of heterogeneous  2.5D / 2.1D / 3D packaging options. We also offer a full range of ASIC design services from our Design Solution Partners as well as directly from Samsung Foundry.

These solutions will allow our Samsung Foundry customers to create new, innovative high performance semiconductors that will drive the next waves of hyperscale datacenter capability by providing the extreme density required to add more functionality, the high speed interfaces to drive throughput, the packaging to deliver highly integrated xPU and near memory combinations, and the ASIC design services to help you put it all together into a winning product.

As you can see, we’re putting a lot of effort into the hyperscale data center market, and I can’t wait to see the results. Perhaps in the not so distant future our collective efforts will enable me to have my very own connected, artificially intelligent R2 Unit.

Learn more about our foundry solutions on our landing page: https://www.samsungfoundry.com/foundry/homepage.do

 

Samsung Foundry Forum

If you’d like to learn more about the industry’s latest innovations in process technology, IP, design tools, design services, and packaging, please consider joining us for the 2018 Samsung Foundry Forum on Friday, May 22, at the Santa Clara Marriott hotel. I look forward to seeing you there!

Tags: , , , , ,

Leave a Reply

Your email address will not be published. Required fields are marked *

Captcha *