接触纳米精密供体器件的全光刻工艺

The ability to fabricate devices with atomic precision holds promise for revealing the key physics underlying everything from quantum bits1,2 to ultra-scaled digital circuits3–7. A common atomic-precision fabrication (APFab) pathway uses a scanning tunneling microscope (STM) to create lithographic patterns on a hydrogenpassivated Si(100) surface8 . Phosphine gas introduced into the vacuum system selectively adsorbs on sites where Si dangling bonds have been re-exposed by patterning9 , yielding atomically precise, planar structures made of P donors. Unlike electron beam lithography (EBL), which can pattern hydrogen with a resolution of around 100 nm and is unable to image the pattern10, the STM is an ideal instrument for this process because it can both pattern and image the hydrogen resist with atomic precision11 . However, STMs are typically capable of patterning devices only up to 10 µm by 10 µm in size, which are too small to directly contact. A post-patterning microfabrication process, consisting of etching via holes in an encapsulating Si overlayer and then depositing metal in direct contact with the planar donor layer, is used to make electrical contact to the devices. Even the largest features made with the STM are small enough that this contacting process relies on EBL for patterning and 200 nm- scale processing. At this scale, making good electrical contact between a deposited metal and an atomically-thin onedimensional line of donors at the edge of an etched hole is challenging, and even successful EBL process flflows in this application are rate-limiting.

In this paper, we detail an all-optical lithography contacting process that reduces the time of fabricating an atomic-precision device by an order of magnitude. This is made possible by the integration of both ion-implanted contacts and metal alignment marks in the starting material, which bridge the scale between the largest regions accessible by STM and the smallest length scale accessible by low-cost photolithography. Specifically, the ionimplanted contacts neck down to a small enough area that the STM can place the APFab device in direct contact with them, and extend out to a region large enough that multiple photolithography steps done to a precision of 2 µm can connect the APFab device to 200 µm sized metal pads. Directly fabricating APFab donor structures on top of ion implanted Si simplifies the burden on microfabrication to that of making contact between deposited metal and an ion implanted region, which can be done with near perfect yield. Moreover, this entire contacting process can be executed in a single day using tools available in most clean-rooms, and can be run on multiple chips in parallel. Moving forward, the increased throughput of our reliable all-optical contacting process promises to dramatically reduce the cost of making new discoveries using APFab devices.

Flashing the sample to 800°C for 5 minutes is sufficient to remove the 10 nm of surface oxide, but leaves the implanted contacts buried in oxide. Since STM cannot tunnel into a thick insulator, the sample must be flashed for a longer period of time to expose enough doped Si to connect directly with hydrogen lithography. Figure 2c shows a topographic image of a sample after it is flashed to 800°C for 15 minutes. Rather than a uniform reduction of all the thick oxide in the implanted region, the partial removal of the thick oxide proceeds from the edge of the implanted region. This leaves a 30-60 nm deep trench that is easily identified by STM. Leveraging the high contrast metal markers and a high-resolution optical camera, we can align the tip precisely to the implanted contacts with sub 2 µm precision. To determine whether the implanted dopants diffuse out of the implanted region, we simultaneously acquired topographic and spectroscopic data using the STM in Figure 2d. This data indicates that there is a region inside the trench that has an enhanced tunneling density of states, corresponding to a high concentration of activated donors16. Moreover, this enhanced density of states is sharply confined to the trench, indicating that As dopants have not diffused out from the implant region.

Fig1

The process of making electrical connection to the APFab device is now simplified as compared to EBL methods, requiring only optical lithography and standard clean-room microfabrication to put metal in direct contact with the eight ion-implanted contacts in a 40 µm by 40 µm area (Figure 3d). Etching down to the implanted Si is complicated by the material stack in that part of the sample, which starts with the Si capping layer, followed by oxide which was incompletely removed during the sample flashing process, and finally by the doped Si. Contacts are made by patterning 2 µm diameter vias with optical lithography followed by a reactive ion etch of the Si capping layer using CF4 at 25°C. Next, the leads are patterned for a lift-off metal deposition. Immediately before metal deposition a relatively long, 90 s etch in 1:6 BOE (buffered oxide etch) is used to remove the remaining oxide in the vias over the ion-implanted regions. After depositing 150 nm of Al, by electron beam deposition, a standard metal lift-off process is used to complete the eight contacts that fan out into bond pads, shown schematically in Figure 3c.

We also examine, in Figure 4b, the transport between contacts that are not connected by a patterned APFab structure. These show a miniscule amount of leakage between isolated contact pairs- less than 0.1 nA at 2 V of bias. Two control samples- one which saw the same thermal processing as our APFab device but no phosphine dose, and a second one which was not subjected to any thermal processing- show similar levels of leakage current to one another. This indicates that the thermal budget of our process does not lead to enough As implant diffusion to be measurable. This also suggests that the additional leakage current between isolated contacts in the patterned sample originates largely from phosphine adsorbing through imperfections in the hydrogen resist. Both the Ohmic conduction through the nanowire, and the small leakage between unconnected pairs of contacts, compare well to an earlier effort which realized metal silicide contacts, but reported nonlinear I-V curves through an Ag nanowire with much larger leakage between unconnected contacts.