General

Written by Steven Woo

In our previous blog post,we discussed the history of neural networks (NNs) and machine learning (ML). We also took a closer look at some of the memory standards that are currently powering a diverse range of NNs and ML applications. In this blog post, we’ll explore some of the more prominent neural networks that are behind the recent advances in AI and ML.

Multilayer Perceptrons

The foundation of modern neural networks is the perceptron, a digital model of the biological neurons found in our brains. Perceptrons are interconnected in layers in a manner similar to how neurons are connected. Multi-layer perceptrons (MLPs) are one of the oldest forms of neural networks, and are used today in many classification tasks where the goal is to identify known objects within a set of data. Convolutional neural networks (CNNs), recurrent neural networks (RNNs) and Generative Adversarial Networks (GANs), leverage this same foundation. As one of most popular neural networks, MLPs continue to be used today for many different types of classification tasks.

As the name implies, MLPs implement multiple layers of neurons, typically 3 to 20, although some have been reported to have as many as 100 or more hidden layers. CNNs add additional layers, known as ‘filters,’ to the front of the network to extract features from the data which are subsequently passed though as inputs to the MLP layers. This technique allows the filters to extract and access features the MLPs use to classify inputs. MLP layers are trained and adjusted with back propagation using mathematical equations to organize the connections and neuron characteristics. It should be noted that the above-mentioned filters are also adjusted using a similar set of equations which are iteratively applied during the CNN’s training phase.

Spiking Neural Networks (SNNs)

Spiking neural networks (SNNs) are a more recent class of neural network that model biological neurons and neural networks more closely than perceptrons and MLPs. Like MLPs, SNNs emulate the neuron and synapses, but they also incorporate the notion of time into the model. In MLPs, each neuron decides to fire at the same time during a propagation cycle. However, SNN neurons do not fire simultaneously during propagation cycles. Rather, SNN neurons are designed to fire at certain times, namely whenever their membrane potential (related to the electrical charge across the membrane) exceeds a specific threshold value.

When an SSN neuron fires, it generates a signal (or ‘spike’) that travels to the other neurons it is connected to. Each of the connected neurons adjust their membrane potential based on this signal – and fire when a specific threshold is reached. There are various methods for using the output signals (‘spike trains’) to encode information, including leveraging the frequency of the spikes or the time between spikes to represent numerical values.

Generative Adversarial Networks (GANs)

Generative Adversarial Networks (GANs) are typically used to power unsupervised machine learning applications. This is a discipline in which machines learn to draw inferences from datasets that are unlabeled, rather than classified or categorized a priori.

Essentially, GANs are implemented by connecting two neural networks using a methodology that allows them to ‘compete’ against each other. The first network, known as the generative network or generator, creates new examples by generating data that fits a data distribution of interest. The second network, referred to as the discriminative network or ‘discriminator,’ discriminates between instances from the ‘true’ data distribution and candidates produced by the generator.

The discriminator, which is often a CNN, is initially trained with a set of learning data until it reaches a specific accuracy threshold. Subsequently, the generator – which is often a variant known as a Deconvolutional Neural Network – creates new examples that attempt to ‘fool’ the discriminator. As new examples are generated and applied to the discriminator, back propagation techniques are used to adjust and optimize both the generator and discriminator.

Interested in learning more about machine learning and AI? You can check out our article archive on the subject here.

Some of the top Rambus technical experts will hold forth at DesignCon 2019 in a full-day sponsored training session on Wednesday, January 30. Plus, the Rambus booth #837 will demonstrate the company’s developments in GDDR6, as well as its comprehensive suite of Ethernet, PCIe, DDR, and HBM IP cores for today’s most challenging data center and networking applications.

We will also be showcasing our CryptoManager Root of Trust and its multiple roots of trust capability for a connected home application.

Included in the technical talks are a half dozen training sessions designed to help audiences understand the latest advances in Rambus’ IP core product line.  They’ll be hearing about achieving better performance and power efficiency in memory systems for AI applications.  They’ll also get the basics of reducing power and cost using 112G XSR technology to enable optical communication adoption.

Other Rambus sessions cover future electrical/optical/signal integrity challenges for wireline communication connectivity, memory options for enabling high-performance applications, and design tradeoffs and system/signal integrity challenges for implementing 112G in FinFET technology. Plus, Rambus will devote one of those training sessions to a secure core implementation with a RISC-V CPU.

Rambus’ training sessions schedule follows:

Memory Options for High Performance Applications

Speakers:  Sreeja Menon, Principal Engineer IP Cores Architecture
Frank Ferro, Senior Director IP Cores Product Marketing
Time: 9:05AM – 9:45AM
Leading edge high-performance computation power and an exploding digital world is limited by memory speed, access throughput, and latency. This presentation discusses the various options available for enablement of a variety of applications, their characteristics, and tradeoffs.

112Gbps ADC Based Long Reach SerDes
Speaker: Ken Dyer, Director IP Cores Architecture
Time:  10:05AM – 10:45AM
The application of high-speed power-efficient ADC’s in long reach wireline communication circuits will be discussed. Design tradeoffs, and System and Signal Integrity challenges for the implementation of 112Gbps in state-of-the-art FinFET technology will be presented.

Enable Innovation with Hardware-Enforced Security at the Core
Speaker: Ben Levine, Senior Director of Product Management
Time:  11:05 AM – 11:45 AM
Making a highly complex processor optimized for performance, into a secure processor is very difficult. Instead, move security-sensitive code into secure hardware cores siloed from general purpose processors, so the cores can utilize processors optimized for secure operation and implement hardware protection against attacks. See an example of a secure core implemented using a custom-designed RISC-V CPU targeted specifically for security; demonstrating a layered security approach offering robust protection against a broad range of threats through carefully thought out system design.

Memory Systems for AI and Leading-Edge Applications
Speaker: Steven Woo, VP, Systems and Solutions and Distinguished Inventor
Time:  2:00PM – 2:40PM
AI and other leading-edge applications achieve higher performance and power efficiency by combining specialized silicon with well-designed memory systems. These systems must address challenging power and physical-design constraints. This presentation will discuss the crucial role that memory systems play and highlight some of the performance and power-efficiency tradeoffs that architects face when designing memory systems for these applications.

Exploring How 112G XSR Technology Enables Adoption of Optical Communication
Speaker: Mondeep Thiara, Senior IP Cores Product Marketing Manager
Time:  2:50PM – 3:30PM
The evolution of optical modules to on-board optics and co-packaged optics. This talk will discuss how 112G XSR technology and silicon photonics can help lower power and cost, allowing a wider adoption of optical communications in data centers.

Next Generation Wireline Communication Interfaces
Speaker: Saman Sadr, Vice President IP Cores Marketing
Time:  3:45PM – 4:25PM
Electrical, Optical and Signal Integrity challenges of the next generation wireline communication connectivity will be reviewed. Solutions for direct connectivity between Electrical and Optical interfaces will be presented, and 2.5D/3D advanced packaging and fabrication technology requirements will be explored.

On November 27^th, Senior Vice President of Global Marketing, Sales and Support, Mike Noonen, will keynote at the SemIsrael Expo 2018, speaking on how “Big Data Needs Secure Tiny Devices.”

Also, on the agenda is Frank Ferro, Senior Director, Product Marketing, Memory & Interface Division, who will cover “Advanced Memory Interfaces for High-Performance Systems,” in the IP & Cores track.  You can visit the Rambus product expert team in booth #26 at SemIsrael.

Semiconductors are becoming more and more important to the future of technology. As an integral part of our tools since the 1950s, they have been powering every computerized device. As software becomes more sophisticated, and as Moore’s Law is fading from prominence, some may wonder how semiconductors can keep up in this increasingly mobile and digitized age. From military hardware in the 1950s, to personal computers in the 1980s, to mobile phones in the 2010s, semiconductors are a vital part of technology, providing the brains for most of our tools.

With everything becoming increasingly connected, from people, to cars, to homes, to cities, to industries, the semiconductor industry is projected to be valued at $1.9 trillion by 2020. While 12 elements of the periodic table of elements were used in semiconductors in the 1980s, that number rose to 16 in the 1990s and 58 in the 2000s. Manufacturers are using more of the periodic table of elements, making much more possible, such as Indium Phosphide optical interfaces and Iridium substrates for diamond wafers.

Semiconductor companies are on the cusp of a fourth tectonic shift in computing. That tectonic shift has offered a set of unique challenges, such as low CPU utilization, where processing capabilities are scaling faster than memory can support, big data analytics, where massive and growing data sets are straining data center architecture, and DRAM scaling, where the cost per bit of DRAM no longer scales with process.

The last 40 years have shown that performance from monolithic scaling is slowing, as increasing transistor counts enable newer functionality and better system performance.  But early systems were bottlenecked by CPU performance, limiting what applications could do. Those bottlenecks have shifted over time from the processor to memory. This is compounded by the rising costs of System-on-Chip (SoC) devices, driven by the increased complexity of equipment and materials required to manufacture said devices.

The question is that if monolithic scaling is not delivering performance and is becoming unaffordable, what can the semiconductor industry do? A passage from Gordon Moore’s 1965 document, where “Moore’s Law” came from, said that “it may prove to be more economical to build large systems out of smaller functions, which are separately packaged and interconnected.”

One thing to consider is system-in-package. Separately packaged, but interconnected devices, known as “disaggregation,” can drive performance and control costs.  Another key technology to consider is high bandwidth memory or HBM2 DRAM. It can output as much as 256GB/s, and provide more bandwidth at 50% lower power with a much smaller footprint than other memory formats such as GDDR5. With big data needing tiny devices, reducing the footprint of memory is important.

The Bottom Line

With the end of Moore’s Law, memory becoming more affordable, and tasks becoming more complex, semiconductors are more relevant than ever.

The financial technology sector converged on Las Vegas for Money 20/20 last week, bringing together the biggest industry names to discuss the future of the payments, finance and commerce.

Following four days of discussions, hundreds of speakers and thousands of attendees, what caught our eye at Money 20/20?

Cryptocurrencies, meet security

The hype surrounding cryptocurrencies and broader blockchain applications remains undiminished, with discussion spanning ICOs, air drops and the seemingly obligatory celebrity tie-in (this year coming courtesy of Akoin).

That said, Asiff Hirji, the chief operating officer of cryptocurrency exchange Coinbase, admitted that the industry was still building to its “iPhone moment”, citing the lack of stability as a leading cause. Given the rampant fraud across the cryptocurrency market, there are few that would disagree.

Fortunately, there are immediately available solutions that provide cryptocurrencies with the security it demands. Tokenization technology, which has already proved successful in securing mobile payments, has huge potential to deliver a frictionless layer of security to complement the immutability of the blockchain.

Securing remote commerce

Jumping to digital payments of a different kind, EMVCo published a draft of its EMV® Secure Remote Commerce (SRC) Specification to help address rising fraud levels in global e-commerce, while reducing consumer friction during online checkout.

According to EMVCo, the draft specification provides a foundation that will enable the processing of e-commerce transactions in a consistent, streamlined fashion across a variety of digital channels and devices, including smartphones, tablets, PCs and other connected devices.

With solutions built on EMV SRC expected in mid-2019, this will be a key trend to watch as momentum for digital payment continues to build. So too will be growing support for EMV card-on-file tokenization. As the merchant stores payment tokens in their database rather than the actual card number, security is enhanced and consumer convenience is increased.

More than ‘just payments’

Following another challenging year for brick-and-mortar retailers across the world, Money 20/20 once again proved that payments innovation and investment is key to creating an enhanced in-store retail experience to compete with online.

Back in 2016, 80% of merchants saw payments as a fundamental part of their business strategy, with 92% expecting to maintain or increase investment over the next 12-18 months (Source: Ovum). Two year after the report, have good intentions translated into meaningful actions?

In short, yes.

Many brick-and-mortar retailers have evolved their offering to streamline and enrich the buying experience. For forward-thinking merchants, mobile wallets are more than just another payment method. Rather, they are a secure platform that enables them to cut waiting lines, deliver real-time incentives and transform rewards programs.

Money 20/19?

As we look towards 2019 and beyond, cryptocurrencies, digital commerce and in-store payments are poised for continued transformation. It is clear, however, that security in concert with simplicity is the killer value proposition. Finding this combination should be the focus of innovation efforts.

For a deeper dive into how tokenization can improve the security of blockchain assets, download the Rambus e-book. To find out more about mobile scan-and-go technology can deliver increased revenues, an enhanced customer experience and optimized operational efficiencies, download the Rambus white paper.

Firefighters are battling three major California wildfires. One is in Northern California in Butte County; another is the Woolsey fire in between LA and Ventura County; and the third is the Hill Fire in Ventura County. Even the town of Paradise in Butte County was virtually burned to the ground.

The New York Times quotes Thom Porter, a Cal Fire Chief, as saying, “There are potentially over 100 people that were killed in Butte.” It is the most destructive wildfire in California history with more than 8,800 structures destroyed, most of them homes, according to the New York Times article.

However, a number of organizations in California are rising to the occasion to provide money and support to aid the thousands of people who have lost their homes or have been displaced. Rambus is among a number of corporations making contributions to bring comfort, assistance, and hope to all deeply affected by these wildfires. Corporate and employee donations from Salesforce, Wells Fargo, Airbnb, Oakland Athletics, San Francisco 49ers and The Golf Relief and Assistance Fund have donated to assist these communities.

At Rambus, we are dedicated to strengthening and improving the quality of life in our communities through partnerships, engagement and employee giving. Our corporate social responsibility is vital to help these communities heal. Rambus is making a corporate donation of $25,000 towards the United Way of Northern California to help support the victims. Moreover, the company is encouraging its employees to donate with Rambus matching every donation of $200 or more.

The company is also making available to its employees other avenues to make donations. Rambus suggests such organizations as the American Red Cross, Caring Choices, Enloe Medical Center, Humane Society of Ventura County, North Valley Community Foundation, and United Way of Greater Los Angeles.

Two of these organizations, Caring Choices, and North Valley Community Foundation, are non-profits located in Chico, CA, near the Paradise/Camp Fire site. Caring Choices organizes volunteers to help Camp Fire victims. Volunteers also care for displaced animals, while volunteer certified doctors and nurses offer medical care. North Valley Community Foundation raises money to support organization sheltering evacuees of the Camp Fire.

As of this writing, fifty known fatalities have been confirmed with several firefighters injured.
Progress has been made to contain these fires. Yet thousands of residents have lost their homes and belongings. These tragedies are leaving a traumatic and devastating effect on these communities.

If you or your company would like to donate, below is a list of organizations for you to consider:

• American Red Cross: provides information, resources for displaced individuals and has updated information on response and needs for both northern and southern CA wildfires.
• Caring Choices: This nonprofit, which is in Chico, Calif., has turned into a hub for organizing volunteers to help those affected by the Camp Fire. Volunteers are assigned a variety of duties, including caring for displaced animals and, for those who are certified doctors or nurses, offering medical care. Caring Choices is seeking monetary donations for its operations.
• Enloe Medical Center: This 298-bed hospital is in Chico, the site of multiple evacuation centers for the Camp Fire. It is accepting donations for patients and families who have been displaced.
• Humane Society of Ventura County: This nonprofit is accepting donations to help animals displaced by the Woolsey and Hill Fires. It is taking in domestic animals, such as dogs, cats and birds, as well as livestock.
• North Valley Community Foundation: This nonprofit in Chico is raising money to support organizations that are sheltering evacuees of the Camp Fire. These could include churches, fairgrounds and community centers, said Logan Todd, a foundation spokesman.
• United Way of Greater Los Angeles: This local branch of the national organization is raising money for those affected by the Woolsey and Hill Fires, specifically to help low-income residents.
• United Way of Northern California: This local chapter of the national nonprofit has established a disaster relief fund to offer emergency cash and help to people who have lost their homes, according to a news release.

To learn more about Rambus’ corporate social responsibility within our communities, visit rambus.com/corporate-social-responsibility/.