Datenspeicherung | Future Markets Magazine

Data storage on edge devices need robust and increasingly powerful memory chips. Flash has been a proven solution for many years. But new technologies and materials offer the promise of even faster and more durable memories.

Suitable data memories are also needed for processing data on edge devices. But the memory chips must overcome particular challenges when it comes to storing the data on the devices.

First of all, they need to offer a combination of low cost and high performance. In addition, the storage media should be energy-efficient so that they can also be used in battery-powered devices. Edge devices must be able to withstand lower and also higher temperatures than is necessary with typical PC applications. They have to be impervious to vibrations and shocks too when used in mobile devices. Classic hard disk drives (HDD) are too sensitive in this regard – and also too large for most edge applications.

Flash is proven but pushed to its limits

Flash memories, as familiar from USB flash drives for example, have already proven their worth. These are semiconductor memories in which information (bits) is stored in a memory cell in the form of electrical charges. Whilst flash memories are still somewhat more expensive per gigabyte, they do offer considerably faster data processing processes than HDDs. The read/write speeds that can be achieved in a flash memory are four times higher than those with HDDs. Moreover, they offer a higher density. Thus saving space and weight, and are less susceptible to errors owing to the lack of moving parts.

The development of flash memories has focused in recent years on reducing the size of the cells. And therefore increasing the data density. However, a physical limit has already been reached here with the present-day nanometre structures. And the lifetime and reliability of the memories is suffering as a result. 3D-NAND memories represent one solution. These are memories where – in simple terms – the planar memory cells are stacked vertically in multiple layers. Because of the shorter connections between the memory cells, the storage capacity and storage speed can be increased. And the power consumption reduced, too.

“To respond to the impressive demand for higher storage capacity and reliability while lowering cost per bit, 3D-NAND manufacturers implement innovative techniques”, explains Belinda Dube, Cost Analyst at System Plus Consulting. “They change the storage type, the memory cell design and stack more layers with each generation to increase bit density and hence reduce the die size. The technological changes in the cell architecture and modification of the fundamental memory functions add complexity to the manufacturing process. However these techniques do lower the cost per gigabyte.”

Data storage in the future

Even the current 3D-NAND memories are increasingly reaching their limits. However, as applications become more complex, as the flood of data continues to grow and demands intensify rapidly.

“Data-storage technology has reached a scaling limit. We need new concepts in order to store the data volumes we will produce in the future”, explains Peter Zalden, Scientist at the European XFEL. XFEL operates the largest x-ray laser in the world. Together with researchers from the University of Duisburg-Essen, Zalden used this laser to investigate how data storage could become better and more efficient with new phase-change materials.

Phase-change memories store data by changing the aggregate state of the bits between liquid, vitreous and crystalline. An electromagnetic field, heat or light pulses switch back and forth between the phases. These two different states correspond to the 0 and 1 in binary code. Corresponding memories have the potential to be a thousand times faster and significantly more durable than existing flash memory chips and could allow future generations of smartphones to offer a higher storage density and greater energy efficiency.

Another promising technology is based on magnetic memories. Spin is used in this case, which is the intrinsic angular momentum of the particle. “The spin is closely related to magnetism. It can be impacted by magnetic fields”, says Dr. Viktor Sverdlov from the Institute for Microelectronics at the Vienna University of Technology. “In the same way that information can be stored by applying different electrical charge at specific points, information can also be stored by ensuring different spin at specific points.”

In contrast to flash memories, such memories can be written any number of times and allow very short write and read access times. However, these new storage technologies are still too expensive to be used on a mass scale in edge devices.

What are Human-Machine Interfaces? Data processing and storage takes place directly in the devices on site. As a result, in many cases, a Human-Machine Interface is also required in or on the edge device. Not only does this allow the recorded data and computation results to be displayed. Rather the edge devices can be controlled too. Thanks to increasingly powerful processors and AI, not only can complex graphics be displayed today. Also instructions can be transmitted by means of gestures or voice.

What are Human-Machine Interfaces?

Each person uses a Human-Machine Interface (HMI) several times every day. From the button on the coffee machine to the control knob on the dryer. Touch displays, in particular, have enjoyed widespread popularity for many years. Being used widely in smartphones and now also increasingly in other devices – from home appliances to cars. Just touching the screen activates actions and allows programs to be controlled.

Operation via graphical interfaces

Edge devices can also require such user interfaces; ultimately this is what a smartphone is. As processors become more powerful and affordable, these devices can be equipped today with complex graphical user interfaces. Those interface allow, together with colour and touch displays, convenient use of the device. They also present the data collected by the device or the results of data processing clearly and vividly. All major semiconductor manufacturers meanwhile offer microcontrollers or SoCs (System-on-Chips) for such graphics applications.

A new class of embedded HMI is currently emerging, via which compact operating software is loaded on an intelligent edge device. Such as a smart meter, an intelligent drive, a special controller or another component. The data and control elements are not shown on a separate display in this case. However, rather are viewed with the aid of a smartphone or tablet.

Voice Interface

Many homes already have a digital assistant. Although voice recognition with systems such as Alexa and Siri is still carried out in the cloud. Since the corresponding analysis processes are highly complex, generally require AI systems and need appropriately large amounts of processing power. The disadvantage of processing voice commands in the cloud however is the delay between the spoken command and the response. Even if this delay is minimal. Interaction between human and edge device should be as intuitive, natural and user friendly as possible. Which is why voice recognition will migrate in future to edge devices, where the appropriate hardware resources are available thanks to new AI chips and storage. Such minimisation of latency with voice control allows much more natural interaction with the device.

But there are other reasons too for moving voice recognition to the edge. “The next generation of voice interfaces will process data locally. Because this is the best way to build a trusted, transparent and intimate relationship between humans and devices”, said Joseph Dureau with conviction. He is the CTO of the French start-up Snips.

The company has developed a voice assistant, which runs entirely on the respective device. In other words does not require an Internet connection and does not collect and process user data in the cloud. This is primarily to exclude the possibility of eavesdropping on conversations and ensure that privacy is protected. A private-by-design solution as Dureau describes it. However, it is also to ensure the independence of a cloud connection and the reduction of the data stream.

Manufacturers offer turn-key solutions

Even chip manufacturers themselves have since started to offer special hardware designs. They are supplied with a small form factor, fully integrated software and are fully prepared for production. Such turn-key solutions minimise the time-to-market, risk and the effort involved in development. They enable OEMs to add voice control to their “smart home” and “smart appliance” products without Wi-Fi and cloud connectivity.

However, the user-experience will be stifled if voice-recognition has to be activated first by a wake-up command, to save energy. The semiconductor industry offers solutions in this respect. The first processors specifically developed to run deep-learning algorithms for voice interfaces already exist, in fact. In these cases, the chips are around 100 times as efficient as traditional CPU and DSP architectures. The voice control of an edge device can therefore always be “awake”. Meaning that an activation command would no longer be necessary.

“Always-on intelligent assistants that reside within smartphones and voice-first devices can consume a great deal of power”, said Dina Abdelrazik, Senior Analyst at Parks Associates. “Maximizing battery life on these devices continues to be a challenge for manufacturers. An effective avenue to achieve low power consumption is to focus on efficiencies around components such as the processor, driver or the chip. In doing so, manufacturers have the opportunity to significantly reduce the amount of power required to enable voice processing functionalities.”

Using gestures to control edge devices

Whilst voice-control systems are becoming increasingly more natural, there is no halt to the development of human-machine interfaces for edge-devices. Thanks to 3D depth cameras and sensors, applications will also be controlled in future using gestures or head movements. Or even facial expressions.

For example, Google has already been working for a number of years now in the framework of the Soli project. On detecting movements, gestures and objects in free space, without any need for cameras or sensors. Instead, movements are captured by means of a tiny radar chip. They can track movements at high speed and with high precision in the sub-millimetre range. The chip is so tiny and energy-efficient that it can also be integrated in very small edge devices. Approval by the relevant US authorities had been outstanding owing to the use of the radar frequency band. But was received by Google at the beginning of 2019. It could therefore soon be possible to control edge devices by casually pointing a finger.