A double-donor system in Si has been considered the basic unit for quantum computing, either based on spin states [19
] or charge states [20
]. We study double-donor systems from a more practical approach, i.e., to demonstrate functions such as a dopant-based memory, with one donor as a sensor (conduction path) and another donor as a memory node (trap). If such a system could be identified, it would become possible to design memories in which sensing is done by a single-electron tunneling current via one donor atom, while storage is ensured by an individual donor working as a memory node.
We have focused on identifying signatures of such a double-donor system among our devices, which contain donors randomly distributed in the channel [23
]. The first target was to select devices that exhibit characteristics with a single isolated first peak. As also described in the previous section, a single isolated current peak can be found when only one donor-induced QD controls electron transport, i.e., when the single-electron tunneling conduction path contains only one donor [13
]. When the gate voltage aligns the conduction path donor's energy with the source/drain Fermi level, conduction through the device starts. Characteristics for such a device are shown in Figure .
Figure 3 Single-electron transfer between two donors. (a) Low-temperature ID-VG characteristics showing a single-donor current peak used as a sensor for detecting charging and discharging of a neighboring donor. (b)-(c) Charging and discharging are sensed as abrupt (more ...)
We planned to use this conduction path donor as a sensor for detecting changes of the charge states of nearby donors. For that, we measured ID
characteristics by up-ramping and consecutively down-ramping VG
around the first peak, as shown in Figure . For most devices, changes in the background charges could not be observed. However, for some devices, as the one shown in Figure , we identified abrupt current jumps reflecting sudden changes in potential due to a charging or discharging event. Charging and discharging, observed in consecutive up-ramping and down-ramping sweeps, gives rise to a hysteresis in the characteristics [zoomed-in in Figure ]. This suggests that, within the hysteresis VG
region, the charge state of the trap is different, depending on the sweeping direction. ID
time measurements [see Figure ] show two-level random telegraph signals (RTS), suggesting a two-level trap. In our device, it is most natural to assume that the trap is a donor, either ionized (D+
) or neutralized (D0
). This is because the number of donors in the channel is larger than the estimated number of interface defects. Furthermore, from our previous studies [24
] comparing doped and undoped channel FETs, it is evident that most of the features (irregular peaks) observed in the measured characteristics are due to the channel donors. Our further analysis [23
] reveals that the trap-donor is closer to the front interface as compared with the conduction path donor. We thus identified devices in which two donors work as a sensor and as a memory node, respectively. This allows further investigation of the physics behind single-electron transfer within double-donor systems.
The particular feature of a two-donor system, compared to other single-electron memory proposals [37
], is that each donor can practically only accommodate one electron. Although a second electron could be added to a donor, the energy level for this state (D-
) resides close to the conduction band edge [39
], and it is not expected to be observed under our measurement conditions. In our devices, donors are embedded into a thin Si layer (10 nm) and, in consequence, reside close to the Si/SiO2
interface. For such donors, an increasing electric field shifts the electron wavefunction towards the interface, while still maintaining the electron localization around the donor [16
]. Localization length at the interface is gradually increasing with electric field; in our devices, as VG
is increased, a donor closer to the surface would expand its potential at the interface. This means that the cross-sectional area of the donor QD, seen from the gate, is gradually increasing. In this situation, we suggested that the donor-gate capacitance should be VG
dependent, which allowed us to reproduce in simulation single-electron transfer between the two single-donor QDs [23
]. Using this simulation, we investigated the effects of different donor arrangements on the hysteresis width [23
]. With further progress in dopant engineering, a controlled design of dopant-based single-electron memory devices, working on a principle as described here, could become feasible [40
A wide range of applications can be envisaged when we consider more complex donor arrangements. We demonstrated that an array of several donors, simultaneously working within a single-electron tunneling conduction path, can allow a time-controlled single-electron transfer between source and drain [24
]. In phosphorus-doped nanowire-FETs, an ac gate voltage can change the charge state of the system by exactly one electron. An electron enters the system during the high level of the pulse and leaves the system during the low level. Injection occurs from the source, while extraction occurs to the drain, which gives rise to a single-electron/cycle transfer between the two electrodes. This operation is similar to single-electron turnstile devices proposed with metallic QDs [41
] or with semiconductor QDs [42
], with the key difference that in our devices QDs are dopant atoms. From simulation studies [26
], we found that the natural inhomogeneity of device parameters (mainly donor-gate capacitances) plays an important role in single-electron turnstile operation.
In short, various applications can be designed using charge states of coupled donors, suggesting that there is a rich environment for further research and development of functionalities downscaled to the level of discrete donors.