Fig. 2: Resistance drift, speed, reliability, and endurance of GST467-based superlattice PCM.

a High resistance state (HRS) vs. time, showing low drift in Sb₂Te₃/GST467 superlattice PCM vs. control GST467, both devices with ≈ 40 nm BE diameter. Dashed lines are fit to R(t) ∼ (t/t₀)^v, where v is the drift coefficient, t is the time after programming, and t₀ is a constant. b Drift coefficient v as a function of resistance state for the same superlattice (SL, red symbols) and non-superlattice (blue symbols) devices. c Eight resistance states with low drift maintained > 1 hour in our GST467-based superlattice PCM with 40 nm diameter, enabling a multi-level cell with up to 3 bits. d Effect of fall times on set transition for four types of PCM, as labeled. All pulses have 1 ns rise time and 30 ns widths, and all devices have 40 nm diameter. The minimum fall times to reach the LRS are marked with black dashed arrows. The GST467-based devices can switch with >10x shorter set fall time ( ≈ 10x faster switching) compared to control devices based on GST225, for both superlattice and non-superlattice PCM. Set voltages for Sb₂Te₃/GST467, Sb₂Te₃/GST225, GST467 and GST225 are 0.65 V, 0.8 V, 1.2 V, and 1.3 V, respectively. e Endurance up to 2 × 10⁸ cycles measured for our GST467-based superlattice PCM with 40 nm BE diameter, maintaining a 100× resistance window. f High-temperature HRS stability of our superlattice PCM compared to control devices. After programming to HRS, devices were annealed for 30 min at successively higher temperatures. We reached each of the upper resistance levels by single-shot reset pulses from the LRS. DC resistances are measured back at room temperature after each annealing event. The higher crystallization temperature of GST467 enables higher temperature stability of PCM based on it. The larger HRS ≈ 10 MΩ in Fig. 2f (vs. Figure 1e and Fig. 2e) is due to differences in the amorphous volume originating from the different pulsing schemes. In addition, fabrication-induced variations between devices can also contribute to observed differences in HRS. All resistances in (a-f) are measured with 50 mV dc bias. Devices in a–d and f were well-cycled ( > 5000 cycles) before measurements.