That dramatic price increase has several causes, including surging demand from AI data centers, changing manufacturing standards and supply struggling to keep pace.
All while the common DRAM architecture is relatively simple: a transistor and a capacitor in one memory cell, repeated as many times as required for the size of your (volatile) memory and some additional circuitry.
So, could patent filings point to memory technologies that are both better and easier on your wallet?
The future of memory is likely to be three-dimensional. Since around 2013, companies such as SK Hynix, TSMC and Samsung have been stacking conventional silicon dies in 3D SDRAM system-in-package designs, mainly for AI accelerators and GPUs rather than ordinary PC memory.
DDR6 is expected to introduce support for 3D-stacked DRAM, potentially bringing the technology closer to mainstream PC memory.
A more futuristic development would be monolithic 3D DRAM, in which memory layers are built directly on top of one another. This avoids the challenges of aligning and bonding separate dies or connecting them with vias, although the technology is still a long way from commercial use.
Another alternative may come from an old idea: MRAM (Magnetoresistive Random-Access Memory), now back in the spotlight thanks to NXP. Instead of storing data as electrical charge, MRAM uses magnetic states, combining speed and reliability with the ability to retain data without power.
Whether any of these technologies will make memory cheaper remains to be seen. Patent filings suggest that the next major leap will not just be faster memory, but a completely different way of building it.