Micro Data Centers Enable Big Data and AI for Edge Computing – EE Times Asia

Edge computing plays a vital role in the effective implementation of several embedded applications, such as artificial intelligence (AI), machine learning, deep learning, and the Internet of Things (IoT). However, today’s data centers cannot currently meet the demands of these types of applications. This is where Micro Edge Data Centers (EMDCs) come in.

By bringing intelligence closer to the embedded system, i.e. the periphery, it is possible to create systems with a high degree of autonomy and decision-making capability. In this way, dependency on the cloud (typical of centralized systems) is reduced, with consequent benefits in terms of energy savings, reduced latency and reduced costs.

Autonomous vehicles, robotic surgery, augmented reality in manufacturing, and drones are some examples of early uses of edge computing. To date, the requirements of these applications cannot be met by current data centers (hyperscale, large and co-located) with “cloud services”, which requires supplementing edge infrastructure, such as EMDCs and “periphery services”.

This edge infrastructure, hardware, and edge services must meet the following requirements:

• high computing speed, necessary to process data locally as much as possible (i.e. at the edge)
• high resilience
• high efficiency

An EMDC, born of a collaboration between HIRO and Vicor companies, can integrate edge computing into multiple intelligent applications.

Importance of edge computing

By delegating greater autonomy and decision-making power to intermediate levels, edge computing reduces the response times of critical applications, which can thus operate in real time. The cloud is used as a memory unit for data storage, while data processing is done as much as possible on the edge nodes, also improving data security and sensitivity.

Edge computing technology allows you to dramatically expand the range of possible applications and services, thanks to the ability to support AI natively rather than relying on AI in the cloud.

This approach is particularly suitable for applications such as Industry 4.0, smart manufacturing, 5G, IoT, autonomous vehicles, smart cities, smart hospitals, robotics, machine vision, etc.

HIRO’s EMCC

Based in the Netherlands, HIRO-MicroDataCenters BV specializes in the development of innovative edge infrastructures (hardware and software) capable of delivering intelligent edge as a service to industrial and other end users. Edge computing is an approach that reduces dependency on the cloud by making edge nodes more autonomous, more capable, and more efficient.

Edge computing requires compact, power-efficient solutions that can operate even in harsh environments with space constraints, bringing computing power as close as possible to sensors and other data sources. From a hardware perspective, efficient power systems with high power density and small form factor are required.

Since human contact with edge devices and sensors occurs everywhere, it is necessary to implement the supporting edge infrastructure as a distributed mesh of EMDCs and edge servers, even in places remote locations, harsh environmental conditions and confined spaces. This brings computing resources as close as possible to data producers and users, resulting in very compact form factors with previously unknown high power densities, posing new technical challenges for power efficiency, electrical signal integrity and high reliability.

“Our hardware designs and technology choices strive for excellence in three areas: thermal management, small form factor modularity, and power conversion,” said Fred Buining, Founder and CTO of HIRO.

The solutions developed by HIRO, in particular the company’s EMDC, are scalable and compact edge computing systems that can operate even in confined environments or even outdoors. The EMDC, shown in Figure 1 below, can be configured with a custom combination of any type and number of CPUs, GPUs, FPGAs, and Non-Volatile Memory Express (NVMe) carriers in compact enclosures, allowing to the EMDC to provide power from 1.5 kW up to 500 kW.

EDMC hardware includes a dual-switch fabric (PCIe and Ethernet), creating high bandwidth and configuration flexibility. The switching fabric also allows the creation of large clusters of FPGAs, GPUs and NVMs attached to a single CPU. Made entirely from solid-state components, these rigs require little maintenance and no active cooling system.

As shown in Figure 1, heat can be dissipated through a fan-assisted dry cooler or a completely passive dry cooler, making the EMDC passive. The solution offered by HIRO, compared to similar products, reduces power absorption by 40%.

“We develop small data centers with maximum power up to 5.4 kW, with the size of a shoebox that can be mounted on the wall. It’s an incredible density that doesn’t exist anywhere,” Buining said.

Vicor Power Module

To achieve greater efficiency, HIRO chose to power the EMDC with a voltage of 48 V_CCinstead of the classic 12 V_CC. Widely used in the telecommunications and industrial sectors, this relatively higher voltage has the advantage of reducing I²R losses on the supply network (PDN).

HIRO used Vicor’s high-efficiency, high-density power modules to meet this requirement, providing energy-efficient, high-density power conversion. HIRO chose Vicor’s DCM modules to perform the first conversion from 48V_CC at 12V_CCas shown in figure 2.

Next will come support for chips such as FPGAs, converting from 48V_CC less than 1V_CC at the loading point. HIRO’s EMDCs can also be installed in a solar farm or wind farm, powered by renewable energy sources.

Compared to other solutions based on discrete components or standard power supplies, which are larger and more complex to design, Vicor’s modules offer a reliable and efficient solution with high power density.

Hiro plans to leverage the scalability of Vicor’s power modules to develop a containerized solution based on six racks capable of approximately 200 kW each.

“Many organizations are moving their legacy applications to containerized applications and to a cloud services environment, often in a remote data center,” Buining said. “HIRO began working with early adopters of advanced technology who are looking for a state-of-the-art on-site service environment.”

Building a distributed state-of-the-art infrastructure in university hospitals in Europe is a specific commitment made by HIRO. Hospitals are required to keep data onsite, but to train AI models that can help detect and treat complex conditions like cancer, tumors, and cardiovascular disease, they need large datasets that go beyond their own data.

“We enable medical experts and researchers to effectively collaborate on diverse and distributed datasets and make meaningful advances in cardiovascular disease, cancer, genetic diseases (Alzheimer’s, ALS, Parkinson’s), kidney disease, etc. .by building a distributed, federated and highly secure infrastructure in hospitals,” Buining said.

Without moving or disclosing the data beyond the hospital, HIRO is building the affordable infrastructure that will allow medical experts to train their models using data from other hospitals.

“Edge data centers are also backed by the European Commission, which sees them as a sort of European independence from large hyperscalers,” Buining said. “If we can catch the data at the edge, it doesn’t have to go to the cloud.”

HIRO is also involved in the €16 million (approx. $16.6 million) BRAINE project, which was funded by the EU’s ECSEL Joint Undertaking to the tune of €8.5 million (approx. $8.8 million). Multiple test applications targeting smart cities, smart hospitals, smart manufacturing and robotics, and smart supply chains will be supported by four test bed sites in the EU: two in the Netherlands, one in Italy and a development platform in Hungary.

This article was originally published on EE time.

Maurizio DiPaolo Emilio has a doctorate. in physics and is a telecommunications engineer. He has worked on various international projects in the field of gravitational wave research designing a thermal compensation system, X-ray microbeams and space technologies for communications and motor control. Since 2007 he has collaborated with several Italian and English blogs and magazines as a technical writer, specializing in electronics and technology. From 2015 to 2018 he was editor of Firmware and Elettronica Open Source. Maurizio enjoys writing and telling stories about power electronics, broadband semiconductors, automotive, IoT, digital, energy and quantum. Maurizio is currently editor of Power Electronics News and EEWeb, and European correspondent for EE Times. He is the host of PowerUP, a power electronics podcast. He has contributed to a number of technical and scientific papers as well as a few Springer books on energy harvesting and data acquisition and control systems.