RAMifications: tackling thermal engineering for DDR5

One of the most recent developments is the advent of DDR5 memory, replacing the DDR4 seen so commonly in home PCs. While DDR5 won’t be in the price range of the average consumer for some time, Intel has produced the first desktop processor to support the standard, with AMD expected to follow with a rival processor this year. DDR5 is here, and as with most new hardware, the price should gradually drop until it’s a realistic purchase for the everyday computer user.

The price premium is the result of a marked jump in performance. DDR5 can offer quadrupled die density and dramatically higher clock rates and data transfer rates. From being able to squeeze out more Frames Per Second (FPS) in video games, to offering greater server performance, DDR5 is going to be able to drive huge performance improvements in both recreational and industrial settings.

These extreme speeds are made possible by changes to both motherboard architecture and voltage increases. The voltage for DDR5 will go up to 1.35V from 1.1V, while voltage regulation has been moved off the motherboard and onto the module itself. More power, and more autonomy, can drive a higher level of performance.

This means that many DDR5 modules will run noticeably hotter than DDR3/4. And while that may seem innocuous, it’s hugely important to consider from a design perspective.

Too hot to handle?

DDR5 memory modules will carry their own power management integrated circuit (PMIC) and voltage regulation module (VRM), both of which will pump out heat when running. MSI’s Renesas P9811-Y0 PMIC can hit a surprisingly sweaty 56ºC.

From the perspective of a thermal engineer, this is an interesting challenge. The heat dissipation and the thermal limits of every component within a system – CPU, GPU, hard drives, memory, and so on – will have a knock-on effect on one another. These surprisingly high temperatures for DDR5 change the way we must accommodate heat.

Equipment manufacturers need to ensure that sufficient airflow can pass over the memory modules, so that they can prevent overheating. There is a range of ways to approach this, from liquid cooling via custom-built waterblocks, to in-built cooling systems as part of the DRAM stick itself.

What’s more, every motherboard will be housed in a different device. The cooling demands for a computer case with poor air flow, for example, will be vastly different to the memory slots on a server farm. The size, shape and material of the enclosure will all impact the way in which a design can perform.

It’s easy to see how intimidating this can be. Thermal engineers are faced with an onslaught of variables, and they need to see the wood for the trees – both in terms of the problems posed and the optimum solution.

These engineers need an environment in which they can test different approaches and layouts. This is where thermal simulation comes into its own.

Just like the simulations

Engineers in every field want to avoid investing in costly prototypes for the purpose of testing – particularly in a supply environment still suffering from a global shortage in semiconductors. The need to balance performance/design with the component bill is a test as old as time, and if components are committed to an inferior design, there is no guarantee that more are available.

There is also the pace at which designs can be developed. The supply chain challenges of the last two years have been, to lean upon the tired cliché, unprecedented. Relying upon a supply network that has never faced more challenges, or more strain, is not necessarily the most robust means of pursuing success. The loss of key components can come at a huge opportunity cost.

Thermal simulation is an environment that specifically caters to these obstacles. Having the opportunity to explore customisable designs, in order to determine the layout that would best protect hardware from the thermal impact of DDR5, is hugely useful.

For one, it sidesteps the need for investment in physical components – the properties in digital form are all you need. The ability to simulate as many different designs as you need, across a plethora of different environments and at no component cost, eliminates the risk inherent in the supply chain at present.

By eliminating the need to receive components and then construct designs, simulation also rapidly accelerates the process of testing as a whole. While some more complex designs can take hours depending on the hardware being used, the time taken pales in comparison to the physical equivalent.

All of this ultimately serves to empower the thermal engineer. Giving them a virtual playground within which they can experiment without the restrictions of real-world logistics or expense offers them the time and the energy to invest in delivering the best possible result. For DDR5, and other new technologies, this can only improve their design – and adoption – in the long run.