HP and Hynix - Bringing the Memristor to Market in Next-Generation Memory

The first successful devices are usually the result of a more finicky and expensive set of processes. Also, all the fabrication steps have been ordered, and if necessary had their process parameters pushed out of otherwise optimal zones for the machines in question, to make the desired device come out right. So, after making the first ones they had to find a way to make them with better, commercially viable, yields. Also they had to develop a process flow to make memristors at the same time as, or at least compatible with, the other steps needed for the other devices that will be needed on a marketable chip. It is common to push necessary support circuitry for a device off the initial test wafers (if for nothing else, it helps reduce the variables, if the devices don't appear to work, was it the new device under test or the amp you put right next to it on the chip; better to put another test device in that space, have two to test, and use an off chip amp to test them).

In addition, there may have been issues with the materials used or waste products produced. You can play around with small amounts of more expensive materials in research. Similarly, disposing of small amounts of very nasty waste products can be fine on the research scale, but dealing with the amounts produced after scaling production up might make the venture unprofitable.

Also, many of the process steps may have been far more "hands on" than mass production demands. So all of those have to be automated and process parameters for them optimized.

All of this is based on academic research experience, which, afaik, doesn't constrain itself as much to sticking as close as possible to the established tools as industry research might. But with a whole new class of devices like this, I imagine even an industry research group wouldn't let a little thing like lack of an established, trustworthy tool for a job stand in the way of a promising avenue toward success. I'd say six years is doing very well considering all these issues and many of them are probably still in the process of being ironed out.