The basic characteristics of a PAC based SM from Savransky's work are shown in Figure 1. The characteristics are for a single two terminal device with two threshold voltages. The conducting state, after threshold switching, is characterized by two regions of different slope connected at a transition point marked “T” in Figure 1. The two conducting regions are for the different polyamorphic states.
In operation, starting with the device in the low threshold voltage state (Vth1), it is switched along a load line to its conducting state. The current is then increased to bring the device through the transition point to the higher current region. This moves the material into its second polyamorphic state, the current is then rapidly reduced to zero, leaving the device in its high threshold voltage state. For switching from the high to the low threshold voltage state, the process is reversed. The read operation is to apply a low threshold voltage pulse and detect if the device switches.
Savransky describes a transition step that allows the device in the high threshold voltage state to be switched along a load line directly into the low threshold switch conducting state. I had concerns about this as a possibility, based on the characteristics shown in Figure 1.
If, as is shown in Figure 2, the two threshold switching states are considered as separate devices in the same package--Figures 2(a) and 2(b)--it is difficult at first sight to see how a device in the low threshold voltage state can be directly switched into the conducting state of the high threshold voltage device. (This is illustrated in the figure as the load lines ending in a question mark.) To aid the explanation of the write process for the article, an alternative picture of the write characteristics was developed. It invokes the existence of a quasi-stable dynamic condition for the post switching conducting state for a SM device in both the high and low threshold voltage conditions, as is illustrated in the combined characteristics of Figure 2(c) by the dotted extrapolations. The author of  concurs  with the inclusion of a time dependent quasi-stable extension to the conducting states of the SM device as an aid to understanding.
It appears to me there is another anomaly associated with the switching characteristics of Figure 1 and in the individual characteristics of Figure 2(b), which deserves further brief consideration here. The post switching characteristics of a threshold switch, as illustrated in Figure 3, are usually considered as the sum of two component parts, a constant voltage (Vh), constant current density region (the characteristics of an expanding and contracting filament) in series with an Ohmic resistor.
The value of Vh is determined as the voltage where the back extrapolation of the combined characteristics intercepts the voltage axis. The problem with the characteristics as described by Savransky and Figure 2(b) is the slope of the resistive component of the second state extrapolates to a holding voltage with a negative value. The explanation of this may be as follows and relates to the fact that all of the active material of the SM device is now involved in its conducting state; it has become a bulk device. Some concept of the characteristics of the conducting region when it cannot expand and is constrained by the sidewalls of the structure might perhaps be determined by the experiment illustrated in Figure 4.
It shows the characteristics of the conducting region of a threshold switch, ignoring for the moment the series resistance element. If while in the conducting state, at a fixed current, the filament is subjected to current pulses, shorter in rise time and duration than the time constant of filament expansion, there are three options as the temperature of the material is raised. These options are illustrated for material with a positive, zero or negative temperature coefficients of resistance.
Moving back to the SM characteristics, if the switched material of a PAC SM is constrained by sidewalls of the device and if the material in the conducting state has a positive temperature coefficient of resistance, the resulting dynamic characteristics would extrapolate to a negative voltage axis intercept, as illustrated in Figure 4. Without the actual electrical conductivity as a function of temperature for the PAC materials used in SMs it is difficult to comment further.