Flash may not scale below 10nm - even 19nm is a bit of a stretch - but what can replace it? ReRAM (or RRAM: resistance RAM) now moving into commercial production has several advantages over flash. It is now the leading candidate for global NVRAM domination.
The trouble with flash Flash, like DRAM, stores a bit as an electrical charge. As flash feature sizes shrink - or the number of bits per cell increases - so does the number of electrons storing the bit.
At some point the number of electrons will get too small to be reliably read or stored. Improved signal processing and ECC can help, but these carry their own overhead.
What is ReRAM? ReRAM uses electrically-altered resistance, not charge, to store a bit: high resistance is a 1; low resistance is a 0. Resistance is easier to measure than a cloud of a few hundred electrons.
The resistance is changed by applying voltage, like flipping a switch. One advantage: the required voltage is a fraction of that needed for flash writes, handy for low-power devices.
Several materials can be used for ReRAM. Some offer the endurance - millions of writes - and retention that beat flash, while others offer the speed of DRAM.
Other advantages include:
- Density. Researchers in Taiwan have have shown that standard processes can build tiny ReRAM cells.
- Cost. Manufacturing with fewer and simpler steps than flash requires has been demonstrated.
- Write-life. Some ReRAM can be written millions of times vs. the 10,000 writes of MLC flash.
- Flexibility. Depending on how it is architected, ReRAM can be optimized for density - mass storage - or speed.
- Materials. A variety of materials can be used to create ReRAM. As research continues it is likely we'll see more interesting possibilities emerge.
The ultimate - and perhaps impossible - goal is a device that is as fast as DRAM with the retention of flash.
What about phase-change? ReRAM's biggest competitor is phase-change memory, which uses heat to change resistance. But it appears that the power required to switch states is making PC-RAM uncompetitive with ReRAM.
ReRAM's path is not entirely smooth: flash has huge momentum and many researchers working to extend its life; and the basic physics behind ReRAM are still not well understood. But it looks like ReRAM has all the pieces needed to move ahead in the next 5 years.
Perhaps we'll get that ultimate memory device - fast as DRAM, non-volatile and cheaper than either - this decade.
Comments welcome, of course.