Memory flexibility is key in FPGA design

Although the technology world is mostly focused on the latest cutting-edge technological advances, in fact, a large portion of semiconductor equipment still uses components that are several years old. For example, some of the recent challenges faced by the automotive industry are mainly focused on chips with relatively backward manufacturing processes.

The field of embedded devices that use FPGAs is much broader and deeper than the automotive industry, so it is particularly important for these devices to be able to utilize a variety of different types of components. In some cases, devices that use FPGAs will use some of the latest technology, but in others, they may use chips from 20 years ago.

The choice of memory technology is a particularly important aspect for designers of embedded devices. Many modern FPGAs are designed with a certain amount of SRAM (static random access memory) to facilitate certain algorithms or other embedded software to run within the FPGA. But in many applications, engineers need to connect them with various types of external DRAM (dynamic RAM) to achieve the functions required for specific devices or applications. For example, in image sensor-based devices, real-time AI processing of high-resolution video signals to perform tasks such as identifying different types of actions requires the use of DRAM to implement frame buffering of input signals.

The problem with external memory is that there are so many different types of memory, with a wide variety of performance characteristics, capacities, power consumption, etc. Therefore, it is critical to be able to connect as many different types of memory as possible. The most commonly used one today is LPDDR4 (low-power double data rate 4), which provides designers with a variety of different capacities/densities, speeds, and power specifications to match the specific application they are developing. Due to its low power characteristics, it is particularly suitable for mobile and other battery-powered devices, and can also be used to develop low-power devices in industrial/automotive and applications where thermal management is challenging.

Although the specification for LPDDR4 was initially released in 2014 (LPDDR5 specifications were introduced in 2019), it remains popular with device designers due to its flexibility and lower cost. Given its optimal power consumption, cost, and power life, LPDDR4 is today the most popular DRAM for industrial and embedded systems. In addition, LPDDR4 offers significant improvements in power consumption and performance compared to the earlier released but still commonly used LPDDR2 and LPDDR3 types of memory. As a result, memory companies like Micron offer a large number of different types of LPDDR4 devices to meet the different needs of the market.

As part of its strategy to modernize the small FPGA market, Lattice recently launched the CertusPro™-NX family, the only FPGA family of its kind that currently supports LPDDR4. As a result, engineers can now not only take advantage of faster devices, but also balance low power and low frequency, which is critical for embedded devices. What’s exciting about this combination is that it can bring powerful capabilities to a wider range of devices, such as machine vision and other smart camera-based applications. In particular, AI capabilities can now be brought to smaller, cheaper and lower power devices, expanding the scope of application of this powerful new technology.

From the perspective of memory manufacturers such as Micron, wider support for LPDDR4 in more host or processing devices (such as CertusPro-NX FPGAs) means that the company will have more customers to sell its memory products to. For equipment designers, the addition of LPDDR4 support means that they can benefit from the widespread and long-term application of these mature mainstream memory technologies, as these devices have a long manufacturing life cycle and can obtain long-term stable supply.

Finally, we need to recognize that the world of embedded devices continues to grow and diversify at an astonishing rate. Therefore, when the range of component choices and interconnect methods continue to expand and evolve, it is a true win-win for all parties involved. The combination of LPDDR4 memory and FPGAs clearly enables a wide range of applications that can be developed specifically for a variety of different environments. This is undoubtedly an exciting new era for device designers who want increased flexibility.

Bob O'Donnell is President and Chief Analyst at TECHnalysis Research , a market research firm that provides strategic consulting and market research services to the technology industry and professional financial sectors.