EDACafe Industry Predictions for 2022 – Xfab

Four Reasons for a Positive Outlook on Embedded Memories

By Nando Basile, Marketing Manager for NVM Solutions, X-FAB

The prospects for embedded memories, particularly non-volatile ones, are pretty solid both in the medium and the long term. This is the result of several key technology and business trends, with the macro-scale evolution of modern society and increasing demand for electronics goods both being major contributing factors. The following article gives more detail on the dynamics involved.

The first trend influencing embedded memories is the contingency situation of semiconductor capacity shortage, which leading analysts (such as J.P. Morgan, Gartner and Deloitte) all expect to continue throughout the whole of 2022, and possibly even longer. From a chip availability perspective, customers embedding the required memory IP in their SoC design will be better protected, compared to those having to face the inherent risks associated with multiple supply management – where sourcing separate parts from different vendors and having to comply with SiP integration providers availability present challenges. Embedded solutions, coupled with advanced booking of capacity allocation will result in increased confidence that “everything will be included for the chip to work.” This will mean that customers’ trust in time-to-market delivery and that solutions will be cost effective can be assured, because of optimized utilization of the total silicon required by their application.

The second trend that is being observed is of an epochal nature. It is continued growth in the intelligence required by smart sensors and actuators. Often referred to as ‘smart sensor migration’, today’s applications are increasingly dependent on such devices. SoCs are no longer devoted to the mere sensing/actuating operation, but must be capable of directly processing the data being captured – either to enable on-the-spot adaptive behaviors or to optimize data transfer activities (on a bus network or an RF broadcast) for further processing elsewhere. From a chip architecture standpoint, any smart sensor/actuator will now require an on-board microcontroller, which will in turn need a defined amount of volatile cache (SRAM) and non-volatile memory (mostly EEPROM or Flash). The SoC’s most basic function will be translating the sensor and actuator operations, which are mainly analog, into a digital form and preparing the data derived for transmission by implementing the correct communication protocol. The critical element to take into account is that the semiconductor platform’s required performance parameters are, in most cases, dependent on the sensor application. Therefore, any memory solution must be capable of withstanding the same operating conditions and exhibiting the same levels of reliability as required by the sensors/actuators themselves. It must also ensure a smooth integration with the rest of the application. Embedded solutions are becoming unavoidable. Typical cases include those for the automotive and aerospace sectors – where sensors have to comply with extremely high/low temperatures, possess high-voltage capabilities and even, sometimes, demonstrate resilience to radiation. Similar requirements also extend to other segments where operative conditions may be less challenging, but the application’s reliability still has to comply with the highest safety standards, as it is in the case of medical applications.

All such markets are expected to steadily grow in the future, led by global trends like vehicle electrification, autonomous driving, remote healthcare, etc. In each of these cases, off-the-shelf alternatives to embedded solutions would simply not be cost effective, prove more difficult to manage, or not comply with the operating conditions of the sensor/actuator.

To address the growing need for embedded memory, X-FAB is able to provide flexible turn-key solutions with best-in-class operational resiliency on a consistent platform that is suitable for customers’ sensor/actuator requirements.

The third trend that must be mentioned is also epochal. It is related to the increasing pervasiveness of portable sensor/actuator applications in everyday life, either battery-operated or leveraging energy harvesting. Again, such applications need to be ‘smart’ – with the capacity to translate sensed data into a useful customer experience in the field, or to compress data for delivery over a wired or wireless network (e.g. Bluetooth, cellular, etc.).

Low-power operation will be a paramount requirement for such applications and embedded memories will be the appropriate choice. Embedded solutions will ensure that there is a smooth integration in respect of the power-reduction schemes of the SoC (e.g. gate powering), a shortened interconnect path, plus substantially reduced antenna issues. For leakage and idle-mode power reduction, embedded solutions will be less impacted (or not impacted at all) by parasitic leakages and other drawbacks that customers need to deal with when opting for a heterogeneous approach with multiple-packaged components.

Also, many portable applications will have stringent space constraints to adhere to, thus requiring small form factor components. Take a medical wearable application, for example.  Such applications will very likely set the expectation levels for embedded memory implementation – driving 3D integration where resistive RAM or spintronic memories are encompassed.

A fourth trend, which is now gaining considerable traction and is set to shape our future, is artificial intelligence (AI). In the longer term, this trend will progressively turn the current wave of ‘smart sensing’ into ‘intelligent sensing’, where even more analysis and decision-making autonomy will be done by sensors/actuators in the field, mimicking what human senses and reflexes do. The need to enable new applications or make the existing ones more efficient will require merging the ’smart’ aspect with low power operation, as described above. Image and audio recognition applications will provide the impetus for this transformation, especially when intended for battery-operated solutions. They will be followed by countless new applications that are only limited by our imagination.

When it comes to AI applications deployed in the field (referred to as Edge-AI), silicon implementation will be through different kinds of Neural Network architectures – all leveraging semiconductors arrays and enabling those basic MAC operations at the foundation of the inference and machine learning process. It happens that such arrays are conceptually similar to embedded memory arrays and represent a natural technologic evolution of the current digital arrays (either volatile or non-volatile) to a multi-bit or ‘almost-analog’ operating behavior.

The pivotal point with Edge-AI is that such memory-like networked structures will no longer be ancillary to a CPU, but will turn into the computational center of the whole application. This is in contrast to the conventional microcontroller-based approach, where memory arrays are used to store data travelling back and forth from the CPU (consuming a lot of power and generating unwelcome heat in the process). In Edge-AI architectures it is the array where all this will take place, exactly as the decision-making process in the human brain results from electric paths running across arrays of neurons and synapses.

Since AI is widely recognized as the next big thing to enhance the way we live, following on from the mobile phone revolution, with double-digit compound annual growth rates in the years to come, it is evident that companies possessing a solid understanding of conventional embedded memories and their integration into smart systems will have a real competitive advantage. Some key technology parameters in this arena will be the bit-cell operation scheme, low overall power budget, design flexibility for multiple NVM solutions and the ability to serve many different markets via one technology platform.

In conclusion, we can be very confident that embedded memories have a bright future ahead. They are destined to support the countless applications where sensing/actuating will require great locally situated intelligence.

For such applications, embedded memory solutions will help to alleviate the semiconductor shortage issues in the short term, acting as a key enabler of innovations in automotive and portable applications (in particular for healthcare). They will also pave the way for the proliferation of Edge-AI solutions in the long term.

X-FAB’s expertise and well-defined technology roadmap will present customers with solutions that allow them to differentiate themselves from their rivals. Consequently embedded memories will be implemented that are going to define the next era of our civilization.

About the Author:

Nando Basile is Marketing Manager for NVM and AI solutions at X-FAB. He’s a seasoned professional, with almost 30 years’ semiconductor industry experience, having served in different companies and occupied numerous roles, both technology and marketing related.

www.linkedin.com/in/nando-basile-1980026

nando.basile@xfab.com

Category: Predictions

This entry was posted on Sunday, January 16th, 2022 at 1:00 pm. You can follow any responses to this entry through the RSS 2.0 feed. You can leave a response, or trackback from your own site.