Interpreting the data
Although the actual temperatures have not been measured, nor is it known
how close the “set” pulse is to optimum, the “set” pulse widths (times)
and temperatures that result from the back-extrapolation an shown in
figure 1 are not grossly out of line with expectations. At a first
glance, the results of the back-extrapolation would suggest that the
seeded-bridge model for the PCM set process that was developed in Part 1
is valid and no other types of crystallization processes, such as
electric field or current driven electro-crystallization, are involved
or are playing a significant role in the PCM “set” process.
there are lithographic node differences and a number of structural
differences such as volume of material involved, thermal barriers, and
confining electrodes, the differences in pulse widths of set time ts
=300 ns for Hynix and ts
ns for Samsung are in broad agreement with what would be expected if
the back-extrapolation-based model is valid and higher temperatures for
crystallization are required. The actual operating “set” and “reset”
pulse widths and currents used for the devices on which this analysis is
based are shown as annotations in figure 1.
What then does all of this analysis do to accomplish our main goal, which is to confirm the accuracy of the seeded-bridge model for the PCM “set” step and its link to PCM ETDR data? So far the conclusion is the seeded bridge model will facilitate a PCM design trade-off between: ETDR performance, temperature and set time or set-pulse width. The variable that will link temperature and set time will be the set current. For an optimized set pulse, the slower the crystal growth rates the longer the “set” time. All of which have potential implications with respect to overall PCM memory array performance. As a possible means of further testing the correctness of the assumption of my “seeded-bridge” model for the PCM “set” step and its link to PCM ETDR data, I have used the published crystal growth rates for GST compositions. Kalb  provides results for a number of different GST compositions that are particularly useful—they’re representative of “seeded” crystal growth in that they exclude the delay time for the incubation-nucleation step that precedes growth. This data [Kalb, 8] represents what is called heterogeneous growth, in which nucleation sites already exist (i.e. the material is seeded). Experimentally, these might be the side walls of a container or crystallites; in the case of the modern PCM structure, that will predominantly be the crystallized GST electrode.
Click image to enlarge
Figure 2: Theoretical performance for a PCM device with a 20-nm gap of
amorphous material between upper and lower electrodes based on seeded
growth rates for different compositions.
The three new
lines that have been added to figure 1 to form figure 2 have used the
Kalb growth rates to calculate the crystallization time for a notional
PCM device with a 20-nm gap of amorphous material between upper and
lower electrodes. These new lines are for a fully crystallized PCM
device based on the seeded-bridge model, using an optimized “set” pulse.
For each line, the colored region is where the actual growth rates from
Kalb have been used. The calculated results are close to what would be
expected and although the actual composition of the Samsung material was
not disclosed, there is some degree of correlation between part of the
calculated line for Ge2
and the three data points for Samsung results.