Method of manufacturing a plurality of quantum information processing devices

Abstract

A method (100) of manufacturing a plurality of quantum information processing devices (3) on a substrate (1), the method (100) including: acquiring (104) an atomic scale spatial map of the substrate surface (5); using one or more electronic processors (522), processing the acquired atomic scale spatial map (106) to identify defects (13) in the substrate surface (5), establishing which of the identified defects (13) are target defects and determining the location of the target defects; using the one or more electronic processors (522) to classify (108) the target defects based on their suitability for forming quantum information processing devices, the one or more electronic processors (522) forming a classifier (536), the classifier (536) having been pre-trained with a training data (536a) characterising quantum information processing devices (3); and defining (112) electrical components to control and readout quantum information processing devices (3) formed by at least some of the defects (13) classified as suitable for forming quantum information processing devices (3).

Description

[0001] METHOD OF MANUFACTURING A PEURAEITY OF QUANTUM INFORMATION PROCESSING DEVICES

[0002] The present invention relates to a method of manufacturing a plurality of quantum information processing devices on a substrate.

[0003] The goal of developing a quantum computer is receiving increasing attention because of the significantly enhanced computational power offered by quantum computers.

[0004] In a quantum computer, a quantum bit (also known as a qubit) can encode information. The qubit is similar to a classical bit in that it can adopt a value of 0 or 1, but different from the classical bit in that it can adopt both values simultaneously (in a state known as superposition). The computational power of a quantum computer lies in the superposition. The operation of a quantum computer requires that qubits can interact with one or more other qubits such that the quantum state of one of the interacting qubits cannot be described independently of the state of the other interacting qubits. This process is called entanglement, and qubits that interact in this way are said to be entangled.

[0005] Qubits are implemented in two state quantum mechanical systems (the states corresponding to 0 and 1). For example, qubits may be implemented by trapping photons, trapping electrons, controlling electron spin, or controlling nuclear spin.

[0006] One promising candidate for implementing qubits is using the electron spin of dopant atoms placed in a semiconductor host. In order to address and control the qubits, the dopant atoms must be placed in a precise location. To date, development has focussed on the placement of phosphorous atoms in a (001) silicon surface. The phosphorus / silicon system is chosen because phosphorus has one additional valence electron compared to silicon and produces a spin 1 / 2 hydrogenic defect state that can be used for quantum computing. Phosphorus atoms are also of a similar size to silicon atoms and are thus easily incorporated, and silicon is a well understood system allowing for relatively easy incorporation of phosphorus quantum bit structures with conventional silicon-based electronics. WO 2019 / 210370 discloses a methodology of scanning tunnelling microscope (STM) hydrogen-desorption lithography for incorporating phosphorus atoms in silicon. The methodology comprises the steps of: forming one or more lithographic sites on the surface portion, using an STM tip; dosing, at a temperature below 100 K, the surface portion using posphine gas in a manner such that, a portion of the molecules bond to the surface portion; and incorporating one or more phosphorus atoms in a respective lithographic site by transferring an amount of energy to the dopant atoms. The energy may be transferred by thermal annealing or transfer of energy through an STM tip. The number of dopant atoms incorporated in a lithographic site is deterministic and related to the size of the lithographic site.

[0007] Alternatively, Arsenic dopant atoms can also be used to implement electron spin qubits.

[0008] WO 2023 / 175315 discloses a method of incorporating arsenic dopant atoms in defined locations of semiconductor lattice. The method comprises: (i) forming a passivation layer on a surface of the semiconductor lattice; (ii) selectively removing the passivation layer at one or more incorporation sites to reveal the surface of the semiconductor lattice; and (iii) exposing the incorporation sites to a dopant precursor gas such that dopant precursor moieties including arsenic atoms adsorb to the surface of the semiconductor lattice in at least some of the incorporation sites. Arsenic dopant atoms are incorporated into the lattice at the incorporation sites.

[0009] Like phosphorus, arsenic has one additional valence electron compared to silicon and produces a spin 1 / 2 hydrogenic defect state that can be used for quantum computing. Arsenic atoms are also similar size to silicon to allow their incorporation into a silicon lattice. However, arsenic can be placed with greater accuracy than phosphorus, has a higher incorporation rate at lithographic sites (i.e., is absorbed at more sites) and is incorporated more quickly than other atoms. Arsenic can also be incorporated at room temperature in some semiconductor lattices, such as Germanium.

[0010] There is still a need to develop a method for incorporating large numbers of dopant atoms (arsenic, phosphorus, or other dopant atoms) in a scalable manner that allows for rapid manufacturing of quantum computers with large numbers of qubits (> 100). According to a first aspect of the invention, there is provided a method of manufacturing a plurality of quantum information processing devices on a substrate, the method including: acquiring an atomic scale spatial map of the substrate surface; using one or more electronic processors, processing the acquired image to identify defects in the substrate surface, establishing which of the identified defects are target defects and determining the location of the target defects; using the one or more electronic processors to classify the target defects based on their suitability for forming quantum information processing devices, the one or more electronic processors forming a classifier, the classifier having been pre-trained with a training data characterising quantum information processing devices; and defining electrical components to control and readout quantum information processing devices formed by at least some of the defects classified as suitable for forming quantum information processing devices.

[0011] The method may comprise, prior to acquiring the atomic scale spatial map , incorporating dopant atoms into the substrate surface to form the target defects.

[0012] Incorporating dopant atoms may comprise: exposing the substrate surface to a dopant precursor gas to form a random distribution of target defects over the substrate surface.

[0013] The method may comprise controlling the supply of dopant precursor gas to control the surface density of target defects base and average spacing between target defects.

[0014] Incorporating dopant atoms may comprise: forming a passivation layer on the surface of the substrate; and heating the passivation layer, to generate a random distribution of holes in the passivation layer, the holes forming incorporation sites for dopant atoms.

[0015] Incorporating dopant atoms may comprise: exposing predefined regions of the substrate surface to a dopant precursor gas to form a quasi-random distribution of target defects over the substrate surface, optionally wherein the predefined regions are formed by masking areas that are not to be exposed to the dopant precursor gas.

[0016] Each of the predefined regions may be arranged to incorporate single atoms only.

[0017] The predefined regions may be arranged in an array with regular spacing along at least one direction along the substrate surface. The substrate may comprise a semiconducting material. The band structure of the semiconducting material may vary along at least one crystal direction of the substrate. The spacing between predefined regions may correspond to the variance of the band structure along the at least one crystal direction of the substrate.

[0018] The method may further include, after classifying the target defects, modifying the distribution of defects on the substrate surface.

[0019] Modifying the distribution of defects on the surface may comprises creating additional target defects at defined locations.

[0020] Creating additional defects may comprise placing dopant atoms in an atomic surface layer of the substrate.

[0021] The additional target defects may be created using a scanning tunnelling microscope.

[0022] The scanning tunnelling microscope may also be used to acquire the atomic scale spatial map of the substrate surface. The substrate may be kept under vacuum in a chamber of the scanning tunnelling microscope between acquiring the atomic scale spatial map and generating target defects at defined locations.

[0023] Classifying the target defects based on their suitability for forming quantum information processing devices may include: classifying targets based on their suitability using the identified target defects; and classifying targets based on their suitability if the distribution of defects on the substrate surface is modified.

[0024] Target defects may comprise a controllable quantum property to function as a quantum information processing device or part of a quantum information processing device.

[0025] The controllable quantum property may comprise a property selected from: spin or charge or energy.

[0026] A target defect may comprise a dopant atom. The dopant may comprise a group III or group V elements. For example, the dopant may comprise arsenic or phosphorus or boron atoms.

[0027] Target defects may be established using the one or more electronic processors to classify the types of defect. The one or more electronic processors may form a second classifier, the second classifier having been pre-trained with a training set of images of labelled defects and images without defects.

[0028] The classification of the suitability of target defects for forming quantum information processing devices may be based on the proximity of a defect to neighbouring defects, such that interactions between the defects are suitable for quantum data processing.

[0029] The classification of the suitability of target defects for forming quantum information processing devices may be based on the proximity of an ensemble of two or more defects, such that interactions between the two or more defects are suitable for quantum data processing.

[0030] The method may include: overgrowing the substrate surface to encapsulate the quantum information processing devices formed by target defects classified as being suitable for forming quantum information processing devices.

[0031] The method may further include: after overgrowing the substrate surface; capturing a spatial map of wavefunction overlap of the overgrown surface; and assessing the viability of quantum devices that will be formed by target defects classified as being suitable for forming quantum information processing using a third classifier, the third classifier having been pre-trained with a training set of labelled images and / or data characterising quantum information processing devices.

[0032] The method may further include: defining electrical components to control and readout quantum information processing devices based on the assessed viability.

[0033] At least some of the electrical components to control and readout quantum information processing devices may be formed in the substrate surface. At least some of the electrical components to control and readout quantum information processing devices may be formed in layers above or below the substrate surface.

[0034] According to a third aspect of the invention, there is provided a non-transient computer- readable medium containing instructions which, when run on a system including processing circuitry, cause that system to implement the classifier of the first aspect.

[0035] The inventors of the current application have realised that atomic scale defects in a silicon substrate can be used as a resource for forming qubits and other quantum computing devices. Random and quasi-random distributions of atomic defects, occurring naturally or as the result of patterning techniques, can be utilized for device fabrication. By accurately mapping, controlling, and augmenting the distribution of defects it is possible to achieve atomically precise fabrication with minimal or no direct atomic manipulation.

[0036] The method of the first aspect enables large numbers of qubits to be formed in a silicon- based system in a relatively short timescale. While techniques for silicon are currently less developed than other candidates for quantum computing, silicon is nearly ubiquitous for existing classical computing device fabrication. The method of the first aspect enables existing silicon fabrication knowledge from classical computing to be used with quantum computing devices, making silicon-based quantum computers commercially viable.

[0037] Each device formed according to the method is unique in structure and placement, due to the quasi-random nature of the distribution of dopant atoms or other atomic defects. However, this randomness is used in designing the devices, rather than forcing a regular arrangement / structure .

[0038] Embodiments of the invention will now be described, by way of example only, with reference to the accompanying drawings in which:

[0039] Figure 1 shows a flow diagram of a method for forming a random distribution of qubits in a silicon substrate based on atomic scale defects;

[0040] Figures 2A to 2C schematically illustrates a surface of the semiconductor lattice at the corresponding steps of the method shown in the flow diagram of Figure 1; Figure 3 shows a flow diagram of a method for embedding a dopant atom in a surface of a semiconductor lattice;

[0041] Figure 4 shows a flow diagram of an alternative method for embedding a dopant atom in defined arrangement on a surface of a semiconductor lattice;

[0042] Figure 5 shows a system for performing the methods of Figure 1, 3 and 4; and Figure 6 illustrates a further alternative method for embedding a dopant atom in a surface of a semiconductor lattice, in a random distribution.

[0043] Figure 1 shows a method 100 of forming atomic defect-based qubits in a silicon substrate 1. Figures 2A to 2C schematically show the surface 5 of the silicon substrate 1 at different stages of the method. Figures 2A to 2C illustrate the (001) surface 5 of a silicon lattice in top-down view.

[0044] Although Figures 2A to 2C shows atoms of different elements as being of the same size, it will be appreciated that this is not the case, and the Figures are only schematic.

[0045] Prior to the method being carried out, the surface 5 of the silicon substrate 1 is cleaned. Various method of cleaning a surface 5 of a semiconductor substrate 1 will be known to the person skilled in the art. This may include cycles of sputtering and annealing, and thermal anneals cycled with cooling between them. The cleaning is carried out under vacuum conditions (< 5 x 101 (1mbar). Ex situ pre-cleaning may also include solvent washes with acetone and iso-propyl alcohol.

[0046] Figure 2A shows the prepared surface 5 of the substrate 1.

[0047] In a first step 102, arsenic dopant atoms 9 are incorporated into the (001) surface 5 of the substrate 1.

[0048] In the current embodiment, the arsenic atoms 9 are incorporated at random locations by exposing the surface 5 to an arsenic precursor gas 11 such as Arsine (AsFE). This results in substitutional defects 13 at some silicon atom sites, where silicon atoms 7 in the surface layer 5 are replaced by arsenic atoms 9.

[0049] The surface 5 of the substrate 1 may be exposed to the precursor gas under a vacuum for a predefined time. In general, the pressure and exposure time of the precursor gas 11 is below doses required for saturation of the surface 5 of the lattice 1. For example, the surface 5 of the substrate 1 may be exposed to precursor gas at l x lOlombar for 10 seconds. The time and pressure of exposure may be controlled be such that a desired density of incorporation is achieved.

[0050] Arsine molecules 11 may fully dissociate at room temperature, when adsorbing to a free silicon surface 3. The steps involved in dissociation of the arsine molecule 11 on a free surface are:

[0051] The arsine molecule is first datively adsorbed to the silicon surface, with an adsorption energy of -0.52eV (determined by density functional theory (DFT) calculation;

[0052] In a first reaction step, the datively adsorbed arsine molecule dissociates into an ASH2+ H species by means of an inter-row hydrogen shift. This reaction transfers one hydrogen atom to an adjacent dimer row, producing a hemihydride dimer. The DFT calculated adsorption energy for this structure is -1.85eV. The DFT calculated activation barrier for this reaction step is only 0.23eV, which corresponds to a reaction that is completed on a time scale of nanoseconds at room temperature. The inter-row hydrogen shift is kinetically preferable to the alternative, (and perhaps more intuitive) on- dimer and interdimer hydrogen shifts, which have DFT calculated activation energies of 0.53 and 0.29 eV, respectively.

[0053] The second reaction step is also a hydrogen-shift reaction, which preferentially occurs in the ondimer direction to produce a dimer-end AsH + 2H structure. This reaction transfers one hydrogen atom on the same dimer as the AsH moiety. The DFT calculated adsorption energy for this structure is -2.20eV. The DFT calculated activation energy for this reaction step is 0.63 eV, corresponding to a reaction time scale of milliseconds at room temperature. The alternative reaction, in which the hydrogen atom shifts in the interdimer direction, has a slightly larger activation barrier of 0.72 eV, and is thus kinetically disfavoured.

[0054] The dimer-end AsH + 2H configuration containing a single-valent AsH fragment is a highly transient intermediate corresponding to an extremely shallow minimum on the potential energy surface. In a third reaction step, this species near-instantly stabilizes into the end-bridge AsH + 2H species, in which the AsH fragment bridges between two dimer-ends. The DFT calculated adsorption energy for this structure is -2.62eV. The DFT calculated activation energy for this reaction step is less than 0.05eV The final reaction step transfers a hydrogen atom from the AsH fragment to the silicon dangling bond at the opposite dimer end. This reaction produces the inter-row end-bridge As + 3H. The DFT calculated adsorption energy for this structure is -3.09eV. The DFT calculated activation energy for this reaction step is 0.89 eV; this is the rate-determining activation energy along the full dissociation path. This barrier is still low enough such that full dissociation of an adsorbing AsH3 molecule into inter-row end-bridge As + 3H will be completed on a time scale of tens of seconds.

[0055] Figure 2B illustrates an example of a silicon surface with five substitutional defects 13. As can be seen, the arrangement of the defects 13 is random, and does not follow a regular pattern.

[0056] In order to incorporate the arsenic atoms 9 into the top layer of the substrate 1, the substrate 1 is heated. With the heating, the arsenic atom 9 moves into the lattice, and a silicon atom 7, which has been substituted by the arsenic atom 9, is moved towards the surface. The arsenic atoms 9 are incorporated only into the top layer of the substrate 1. This means the arsenic atoms 9 achieve three bonds to silicon atoms 7 in the top and second layer of the substrate 1.

[0057] The surface density of arsenic atoms is controlled to achieve a target / desired amount by at least the pressure and exposure time of the precursor gas. The desired surface density will depend on the overall design of the device being made. In one example, the desired surface density will be based on the desired spacing between arsenic atoms 9 to form the devices being made. For example, a surface density of I x lO12atoms / cm2will result in an average spacing of approximately lOnm. In general, the surface density is in the region of between I x lO11to I x lO13atoms / cm2.

[0058] In a second step 104 of the method 100, the surface is first hydrogen terminated and an atomic resolution image of the surface 5 is captured using scanning tunnelling microscopy (STM) imaging. The STM image provides a spatial map of features on the surface 5 of the substrate 1. The surface is hydrogen terminated by forming an atomic hydrogen passivation layer. The hydrogen passivation layer is a single atomic layer thick. With the passivation layer formed, all sites on the surface that are not bonded to an arsenic atom 9 are terminated by a hydrogen atom. Various methods for forming such a hydrogen passivation layer are known to the person skilled in the art.

[0059] In a next step 106, the captured image is analysed to map the location of the arsenic atoms 9. In the current embodiment, where the entire surface 5 of the substrate 1 is exposed to the precursor gas 11 and passivated, the arsenic incorporation sites can be mapped by discerning As-Si heterodimers from hydrogen terminated silicon.

[0060] It will be appreciated the surface 5 of the substrate 1 may include other types of defects, in addition to arsenic substitutional defects. This may include, for example, non-arsenic substitutional defects, vacancy defects, interstitial defects, edge defects, planar defects, and Frenkel defects. As part of the step 106 of mapping the location of arsenic atoms 9, the method 100 may include identifying the different defects within the surface and differentiating the arsenic atoms 9 from other defects. In this way the arsenic atoms 9 can be targeted in later steps of the method 100.

[0061] In the next step 108, the target / arsenic defects 13 that have been identified and mapped are classified based on their suitability to be used for qubits.

[0062] Suitability for qubits is judged based on a number of factors. Depending on the device being made, any one or a combination of factors may be used.

[0063] A first factor to be considered is the spacing between neighbouring arsenic atoms 9.

[0064] The entanglement of qubits is based on two or more defects 13 interacting with each other. Depending on the device design, the entanglement may involve interaction between n arsenic defects 13.

[0065] A group of arsenic defects 13 is identified as potentially being suitable when n defects 13 are identified as being separated from each other by less than a first threshold distance dl and separated from all other defects 13 by more than a second threshold distance d2. dl and d2 are based on the coupling mechanism being employed in the qubit. Where n defects are used to make entangled qubits, the n defects should be within a distance of each other to allow the coupling between them and should be far enough from all other defects 13 to avoid unwanted coupling with other defects.

[0066] For example, in exchange coupled qubits, a spacing of around dl ~ lOnm is sought for the defects 13 forming the qubit and a spacing of d2 ~ 20nm from other defects 13 is sought.

[0067] In another example, the spacing may be chosen such that interaction between valence electrons on neighbouring arsenic atoms 9 can be manipulated by an applied electric filed so that the two electron states become entangled.

[0068] The simplest qubit entanglement makes use of an isolated pair of defects 13. In this case, a target / arsenic defect 13 may be identified as suitable if it is separated from its nearest neighbour by a spacing of less than dl and if the defect 13 and the nearest neighbour are separated from all other defects 13 by a distance of more than d2.

[0069] In entangled qubits involving more than two defects 13, it may be that each of the defects 13 forming the entangled qubits is within the first threshold dl of all other defects 13 forming the entangled qubits and spaced from all other defects by at least the second threshold d2.

[0070] On the other hand, in more complex designs more detailed patterns may be considered and it may not necessarily be the case that each defect 13 forming the entangled qubits interact must be within the first threshold of every other defect 13 forming the entangled qubits. This may be the case, for example, when not all defects forming the entangled qubits need to interact with every other defect. For example, in entangled qubits involving three defects 13, it may be that the spacings between the first and second defects 13, and the second and third defects 13 are both less than dl, but the spacing between the first and third defects 13 does not need to meet this requirement. In this case, the requirement that all defects 13 forming the entangled qubits be spaced from other defects 13 by at least d2 is maintained. It will be appreciated that as the complexity of the design increases, the requirements on the spacings may be modified to allow interactions within entangled qubits and between entangled qubits. As the number of defects involved in single entangled qubits / devices increases, the number of neighbours that needs to be assessed for determining suitability will also increase.

[0071] Furthermore, it will be appreciated that the suitability of a defect 13 or group of defects may be graded on a variable scale based on deviation from ideal / optimum spacings whilst still remaining within a desirable range. For example, a pair of defects 13 separated from each other by 7nm and from all other defects by 30nm might be deemed more suitable than a pair of defects 13 separated from each other by 1 Inm and from all other defects 13 by 20nm.

[0072] A second factor used in assessing the suitability of defects 13 for forming devices is the proximity of arsenic atoms 9 to other surface defects. The spacing from other surface defects should be large enough to avoid unwanted coupling interactions.

[0073] In addition to the proximity of a defect 13 to neighbouring defects 13, various other factors may be included. This includes, for example, the space around a defect 13 or group of defects to provide electronic control structures, such as electrodes, where needed. Where electronic control structures are required, this may increase the minimum spacing d2 of defects 13 (or groups of defects 13) forming entangled qubits from other defects 13.

[0074] Suitability of a defect 13 or group of defects 13 for use as entangled qubits is not simply based on the current placement of arsenic atoms 9. The identification of arsenic atoms 9 as suitable for use as qubits is also based on suitability if the distribution of defects 13 on the surface 5 is modified.

[0075] This may be based on the spacings discussed above.

[0076] For example, in a device with entangled qubits made of pairs of defects 13 spaced from each other by less than dl and from other defects by at least d2, then a first (lone) defect 13 spaced from all neighbours by sufficient spacing to provide a second defect 13 that is dl from the first defect 13, with the second defect 13 also spaced from all neighbours by d2 may be identified as suitable.

[0077] In general if a defect 13 or group of defects 13 would be useful for entanglement if an additional defect (or defects) was placed near them, and there is space in the defect distribution to allow such placement while maintaining the requirement of all defects in the group being spaced from other defects 13 by at least d2, this defect 13 or group of defects 13 will be identified as suitable.

[0078] In step 110, the surface distribution of defects 13 is modified using the STM used for capturing the image in imaging step 104.

[0079] During the modification step 104, additional arsenic atoms 9 may be provided at locations identified in the previous step 108. It will be appreciated that any number of arsenic atoms may be added.

[0080] Newly added atoms should be provided at precise locations to ensure a suitable ensemble 15 is formed. Additional atoms 9 are placed using a hydrogen resist lithography process, such as disclosed in WO 2023 / 175315, which is incorporated by reference. The hydrogen resist lithography may make use of the passivation layer formed previously when generating the atomic resolution image of the surface 5. Alternatively, a new resist layer may be formed.

[0081] Figure 3 illustrates a method 200 of a hydrogen resist lithography process for precisely positioning arsenic atoms 9, after formation of the resist layer.

[0082] At a second step 202 of the method 200, the tip of the STM is used to selectively remove hydrogen atoms 11 from the passivation layer, to expose the surface 7 of the substrate 1. The removed atoms define incorporation sites where additional arsenic atoms 9 are to be incorporated.

[0083] The precise location of the STM tip is controllable through the control routines of the STM. By controlling the position where the incorporation sites are formed, the location of the arsenic atoms 9 can be controlled. The size of the incorporation site may be varied. For example, the incorporation site may have an area of four silicon atoms 7, preferably arranged in a square (i.e., two adjacent silicon dimers) providing four dangling bonds. Alternatively, each incorporation site may be the size of a pair of silicon atoms 7 (a silicon dimer or two silicon dangling bonds), or a single silicon atom 7 (a single dangling bond).

[0084] In a third step 204, the incorporation sites are exposed to arsine gas 1 1. In one example, the incorporation sites are exposed to Arsine gas at 5 x l0-9mbar for 5 minutes, although it will be appreciated that this is by way of example only, and a range of suitable pressures and times may be used.

[0085] As discussed above, arsine molecules 1 1 may fully dissociate at room temperature, when adsorbing to a free silicon surface 3 without defined lithographic sites, or with sufficiently large lithographic sites. For lithographic sites having a size of four of fewer dangling bonds, the dissociation mechanism is halted partway along the reaction path, with an AsHxmoiety bonded to the lattice. This is due to the lack of availability of dangling bonds for intra- or inter-dimer transfer of hydrogen atoms due to the passivation layer. Therefore, various AsHxmoieties (where x = 0 to 3) are adsorbed onto the silicon surface 5 at the incorporation sires. Away from the incorporation sites, the passivation layer 9 prevents stops the moieties 17’ from adsorbing.

[0086] At the incorporation sites, the moieties adsorbed will depend on the size of the incorporation site 15. Depending on the AsHx moiety, the arsenic atom 5 may bond to a single silicon atom 7 (referred to as an end configuration), or across the two silicon atoms 7 at the incorporation site 15 (referred to as a bridging configuration). At incorporation sites where the AsHx moiety bonds to a single silicon atom 7, the dangling bond at any other silicon atoms 7 at the incorporation site bonds to a hydrogen from a dissociated arsine molecule 1 1. The table below, which shows the percentage termination of dangling bonds at incorporation sites of different sizes: Where x = 1, the AsH moiety bonds in an end configuration. Where x = 2, there is a mixture of moieties bonded in the end and bridge configurations.

[0087] Typically, two arsenic atoms 9 are only seen at incorporation sites having an area of six or more silicon atoms 7. Therefore, a single arsenic atom 9 can be placed by using incorporation sites having an area of fewer than six silicon atoms 7.

[0088] In a fourth step 206 of the method 200, the surface 5 is annealed to incorporate the arsenic dopant atoms 9 into the lattice structure, away from the surface 5. The annealing is performed by heating the sample to 350°C for approximately 1 minute.

[0089] The annealing step 206 moves arsenic atoms such that they become incorporated into the surface 5, at the site of adsorption, in both the first step 102 of the method 100 (the initial exposure of the surface 5 to arsine 11) and arsenic atoms incorporated into the surface 5 during the surface modification step 110.

[0090] It will be appreciated that the annealing temperature may be chosen over a range of temperatures. The minimum temperature required is dictated by activation temperature of diffusion and desorption of hydrogen atoms in the passivation layer. Incorporation of arsenic atoms 9 requires enough free silicon sites (dangling bonds) are available for full dissociation if the arsenic moieties, as will be discussed below. These free sites are provided through the anneal, by diffusion of one or two dangling bond windows to the adsorption site, and / or by thermal desorption of the H atoms. The minimum temperature required for hydrogen desorption is approximately 400°C, for diffusion of a single dangling bond is 275 °C and for diffusion of two dangling bonds is 375 °C. The maximum temperature is determined by the activation temperature of recombinative AsHxdesorption, which is around 450°C. Therefore, the anneal may be between 350°C (to achieve hydrogen desorption) and 450°C (to prevent desorption of arsenic)

[0091] It will also be appreciated that the time of the anneal, and heating ramp up and ramp down rates may also vary.

[0092] During the anneal, vacancies to complete the dissociation of the AsHxmoieties are generated by diffusion or desorption as discussed above. As also discussed above, the barrier to creating H-vacancy delivery is lower than that of AsHxrecombinative desorption. Therefore, a high proportion of the arsenic atoms 5 are incorporated.

[0093] The below table illustrates the incorporation rate for different size incorporation sites (i.e., the percentage of arsenic moieties bonded to the surface 5 which result in an incorporated arsenic atom 9):

[0094] As discussed above, when an arsenic atom 9 is incorporated into the surface 5, the arsenic atom 09 moves into the lattice structure, and a silicon atom 3, which has been substituted by the arsenic atom 9, is moved towards the surface. The arsenic atoms 11 are incorporated only into the top layer of the silicon 1. This means the silicon atoms 5 achieve three bonds to silicon atoms 3 in the top and second layer of the lattice.

[0095] In the method 200 discussed in relation to Figure 3, the position of the arsenic atoms 9 is known to within a few silicon atoms 7 in the lattice, depending on the size of the incorporation site.

[0096] In general, the larger the incorporation site, the higher the proportion of sites are successfully doped, but the lower the accuracy the positioning of the arsenic atom 9 is known. For example:

[0097] For an incorporation site of four silicon atoms 7 in area, single arsenic atoms 9 are incorporated at all but a negligible number of incorporation sites, and the position is known to within a four-atom area.

[0098] For an incorporation site of two silicon atoms 7 in area, single arsenic atoms 9 are incorporated in approximately 80% of sites, and the position is known to within a two-atom area. For an incorporation site of one silicon atom 7 in area, single arsenic atoms 9 are incorporated in approximately 10% of sites, and the position is known to within a single atom.

[0099] Figure 4 illustrates an alternative method 300 that can used to provide even greater accuracy in the positioning of the arsenic atoms 9. The method 300 of Figure 4 is the same as the method 200 of Figure 3 unless stated otherwise.

[0100] As with the method of Figure 3, the resist layer used for atomic resolution image of the surface 5 may be used in the lithographic process, or a new resist layer may be formed.

[0101] According to the method 300 of Figure 4, the formation of incorporation sites 303 and exposing to arsine gas 304 is the same as in the corresponding steps 202, 204 in the method 200 shown in Figure 3. However, in the method 300 shown in Figure 4, the area of each incorporation site is only a single silicon atom 7 in the lattice structure (i.e., a single dangling bond).

[0102] As discussed above, for an incorporation site of a single silicon atom 11 or a single silicon dimer, arsine moieties are not adsorbed at all sites. After the step 304 of exposing the incorporation sites to arsine gas 11, the method 300 of Figure 4 includes the step 308 of identifying, using the STM tip, the incorporation sites where no moieties have been adsorbed.

[0103] In a next step 310 a check is carried to determine if an end criterion for the method 300 has been reached, as will be discussed below in more detail.

[0104] If an end criterion has not been reached, the method 300 reverts to the second step 302.

[0105] At sites where the arsine moieties are not absorbed, dissociated hydrogen from the arsine molecules will have filled the incorporation sites by terminating the dangling bonds.

[0106] When the method 300 moves from the check step 310 to the step 302 of removing the passivation layer at incorporation sites, the passivation layer is only removed at the incorporation sites that have been identified as not already having AsHxmoieties adsorbed. Sites which have been determined to have moieties adsorbed are not exposed. The adsorbed moieties prevent further adsorption in later stages, and thus now form part of the passivation layer. Due to the energy of the bond between the moiety and the silicon surface 5, the adsorbed moiety will not desorb from the surface and recombine.

[0107] The loop 312 of forming incorporation sites 302, exposing to arsine gas 304, identifying incorporation sites without arsine moieties 308, and checking the end criteria 310, is repeated iteratively until the end criteria is met. Then a thermal anneal is performed at step 306 to incorporate the arsenic atoms 9 from the absorbed moieties. As discussed above, approximately 100% of adsorbed moieties result in absorption of an arsenic atom 9.

[0108] Various end criteria may be used in the checking step 310. These include, for example: 100% adsorption of arsenic moieties at incorporation sites (i.e., no empty sites identified);

[0109] The percentage (or number) of incorporation sites with adsorbed arsenic moieties is above a threshold (where the threshold is below 100%);

[0110] A fixed number of loops have been completed. The number of loops may be determined based on a statistical likelihood of reaching 100% adsorption of arsenic moieties at incorporation sites (or the proportion of incorporation sites with adsorbed moieties being above a threshold);

[0111] A fixed number of loops being completed without any change in the number of sites without arsenic moieties absorbed; or

[0112] A fixed time.

[0113] Other suitable end criteria may also be used.

[0114] By the method shown in Figure 4, high levels of incorporation of arsenic atoms 9 can be achieved with precise placement of the arsenic atoms 9, such that the position in the arsenic atoms 9 is known to within a single atom in the lattice when using incorporation sites of a single atoms, or to within two atoms in the lattice when using incorporation sites of a single silicon dimer. As with the method of Figure 3, it will be appreciated that the steps 300 of Figure 4 may be automated.

[0115] It will be appreciated that the iterative loop 312 need not only be used in combination with incorporation sites having an area the size of a single silicon atom 3 or dimer. Where a high percentage of dopant incorporation is required (for any dopant, not just arsenic), the iterative process may be used to ensure the number of incorporation sites with no dopant is reduced.

[0116] It will be appreciated that the greater the proportion of moieties adsorb in each loop, the fewer loops will be required to achieve a high level of overall incorporation. For example, where the incorporation sites are a single dimer in size, 80% of sites have adsorbed moieties in a single loop, and so far fewer loops are required than when the incorporation sites are a single atom in size, when only 10% of sites have adsorbed moieties in a single loop.

[0117] It will be appreciated that in some cases, incorporation sites may be increased in size in later loops, so that an end criterion may be achieved.

[0118] Figure 2C shows the surface of Figure 2B, with an additional substitutional defect 17 to form the desired ensemble 15.

[0119] After or at the same time as the modification of the distribution of defects 13, the method 100 of Figure 1 includes the step 112 of defining electrical components (not shown) such as electrodes in the same layer as the arsenic atoms 9.

[0120] This may include forming highly doped regions (not shown) on the substrate 1 (with the same or different dopant used to form defects 13), or deposition of other conducting materials, such as gold. The electrical components include electrodes, control gates to mediate interaction between arsenic atoms 9, contacts for providing signals and reading out signals and other structures for controlling the qubits in the final devices and for assessing the exact qubit separations in later steps 116 of the method 100.

[0121] It will be appreciated that electrical components may also be formed above or below the plane in which the arsenic atoms 9 are incorporated, as will be discussed below.

[0122] The electrical components may be formed by the same hydrogen resist lithography process as discussed above, by a separate hydrogen resist lithography step, or by any other suitable method. In a next step 114, the surface layer 5 with embedded arsenic atoms 9 is encapsulated by overgrowing a layer of silicon. The overgrown layer may be between 3 and lOOnm thick.

[0123] In a subsequent step 116, the method 100 makes a further assessment of the viability of defects 13 or ensembles of defects 15 as qubits. This can be done in a number of ways. In a first embodiment, atomic force microscopy (AFM) with electron sensitivity may be used to make measurements of the qubits. This may optionally make use of the electrical components defined in the previous step 112.

[0124] Instead of or as well as use of AFM, atomic scale modelling may be used.

[0125] The AFM and / or atomic scale modelling provides a spatial map of charge states of valence electrons of dopants, from which the electron wavefunction overlap of potential qubits may be determined. This also provides a spatial map of charge trap defects which may be formed during the overgrowing step 114.

[0126] In a further example, a classifier may be used to assess the viability of qubits. As discussed in more detail below, the classifier may be trained on a database of useful separations for forming entangled qubits.

[0127] In further steps 118, additional device fabrication techniques may be used to complete and connect the devices identified as viable in step 116. This may include patterning further electrical components and control structures in various layers, and other suitable processes.

[0128] During the method 100, the substrate 1 is kept under vacuum until at least the step 112 of defining electrical components in the same layer as the arsenic atoms 9 is completed. These steps are performed under vacuum conditions. The vacuum chamber in which the steps 102-112 are carried out may be arranged to carry out cleaning, arsenic deposition, hydrogen resist lithography and STM processes in a single chamber or multiple chambers, as discussed below.

[0129] In one example, the substrate 1 may be kept in the same vacuum chamber for the step 114 of overgrowing and the step 116 of assessing viability using AFM. In other examples, the step 116 of assessing viability using AFM and the step 114 of overgrowing may be performed separately (although again under vacuum conditions).

[0130] When the overgrowing step 114 and AFM step 116 are carried out at a separate location, a temporary passivating surface, such as a resist layer, may be applied before the substrate 1 is removed from the vacuum chamber used for the first part of the method (up to the step 112 of defining electrical components in the same layer as the arsenic atoms 9).

[0131] The further processing steps 118 completed after over growing 114 can be performed separately and can be performed before or after the AFM step 116. In this case, the overgrown surface acts to protect the devices embedded in the lattice.

[0132] Figure 5 schematically illustrates a system 500 for carrying out the method 100 of Figure 1 (and optionally the methods 200, 300 of Figures 3 and 4).

[0133] The system 500 includes a controller 502 for controlling operation of an STM 504, an AFM 506, and for performing the analysis steps required in the above methods 100, 200, 300.

[0134] It will be appreciated that in the example shown in Figure 5, all steps are controlled by a single controller 502 and carried out in the same system 500. However, this is by way of example only, and the functionality of the controller 502 may be split between two or more different systems.

[0135] The controller 502 includes a first driver 508 for controlling operation of the STM 504. This may control various functions including image capture, and manipulation of the surface 5 of the substrate 1 where the STM is used for modifying the surface distribution of defects 13 and / or hydrogen resist lithography processes.

[0136] A second driver 510 is provided for controlling the operation of the AFM 506. The AFM driver 510 controls similar operations to the STM driver 504.

[0137] During processing, the substrate 1 is provided in a controlled vacuum chamber 514. It will be appreciated that the vacuum chamber 514 may have different regions or sub- chambers 514a-c in which different stages of processing may occur. For example, separate chambers 514a-c may be provided for interaction of the sample with the STM 504, with the AFM 506, for deposition of arsenic, or hydrogen resist, for cleaning of the surface 5 of the substrate 1 and for other processes.

[0138] Further drivers 512a...n are provided for controlling various other functionalities within the chamber 514. This may include, for example, operation of vacuum pumps (not shown) to form or maintain a vacuum in the whole chamber 514, or one or more sub chamber 514a-c, movement of the substrate 1 between sub-chambers 514a-c, controlling deposition of hydrogen resist, controlling exposure of the surface 5 of the substrate 1 to precursor gas 11, cleaning processes and the like. Automated operation of such processes will be known to the person skilled in the art.

[0139] The controller 502 also includes a memory 516 having a programme storage portion 518 and a data storage portion 520. The memory 516 may be non-transient. The programme storage 518 comprises computer programme instructions that cause operation of the controller 502. The separate computer programme instructions may be considered as separate modules. The data storage portion 520 contains various reference data, and other stored information as required.

[0140] The controller 502 also includes a processor 522 arranged to execute the computer programme instructions stored in the programme storage portion 518 of the memory 516 to cause the operation of the modules defined therein.

[0141] A user input / output interface 524 is also provided, for receiving user instructions / input and providing data output to a user. The user input / output interface 524 may be a single unit, such as a touch screen device, or may comprises multiple units, such as a display, projector, keyboard, mouse and the like. The user input / output interface 524 may also include a communications interface such as wired or wireless communications to allow for remote operation. This may include, for example, 3G, 4G, 5G, WiFi, Bluetooth, local area networks, wide area networks, the internet and the like.

[0142] The memory 516, drivers 508, 510, 512, input / output interface 524 and processor 522 are all connected to each other through a system bus 526. As discussed above, the programme storage portion 518 includes various modules to control operation of the system 500.

[0143] An STM module 528 may be provided to control operation of the STM 504 based on inputs provided by a user or another programme module. Similarly, an AFM module 530 may be provided to control operation of the AFM 506 based on inputs provided by a user or another programme module.

[0144] The STM module 528 and AFM module 530 may make use of respective calibration data 528a, 530a to allow for precise control and positioning.

[0145] The programme storage portion 518 may also include modules 532, 536, 540 to implement various machine learning classifiers as required by the method 100 of Figure 1.

[0146] Any suitable type of classifier may be used for the classifiers 532, 536, 540 discussed herein. For example:

[0147] An artificial neural network (ANN), for example a feed-forward ANN. It will be appreciated that any suitable arrangement of layers and nodes may be used;

[0148] A Logistic Regression Model (LRM);

[0149] A Naive Bayes algorithm;

[0150] A K-Nearest Neighbours (KNN) classifier; or

[0151] A Support Vector Machine (SVM).

[0152] For any of the above classifiers 532, 536, 540, the classifier 532, 536, 540 undergoes supervised training using labelled training data. The supervised training modifies the weights, coefficients, and other parameters of the classifiers to achieve the desired outcomes. The training data used for each classifier 532, 536, 540 will be discussed below in more detail.

[0153] A first classifier 532 is used in the step 106 of mapping the location of substitution defects 9. In one embodiment, the first classifier 532 identifies all features on the surface 5 of the substrate 1 in separate classes. This may include, for example, silicon atoms 7, arsenic dopant atoms 9, other types of dopant / impurity, different types of defects and the like.

[0154] In other embodiments, fewer classes may be used. To do this, certain types of defects / features may be grouped into a single class. For example, all edge defects may be in one class, all planar defects in another class. In another example, arsenic atoms may be in one class, and silicon atoms in another class, and all other dopant / impurity atoms in another class. In an extreme example, all features may be classified on a binary level of arsenic atom or not arsenic atom.

[0155] The output of the first classifier 532 contains information on the location of each arsenic atom in a frame of reference that can be used for later processing. This may be provided in a datafile 534 stored in the data storage 520 portion of the memory 516 or in any other suitable form. In some examples, the output of the first classifier may simply be provided to the second classifier 536 without being stored.

[0156] The training data for the first classifier 532 comprises a set of images with the features labelled with the different classes. The labelling is applied by an operator. The images are captured under similar conditions to the processing of the substrate 1 discussed above. The images in the training data are of (001) surfaces of silicon lattices that have undergone the same processing steps as will be undergone by lattices 1 in the method.

[0157] In one example, the training data for the first classifier 532 comprises includes five 100 x 100 nm2STM images with 512 x 512 pixel resolution. The images are manually cropped and labelled giving a set of images with arsenic dopant atoms 9 and a set of images without arsenic dopants atoms 9. 20% of the training data is held back for validation once the classifier 532 is trained.

[0158] In one example, the training data may be stored as a datafile 532a in the data storage 518 portion of the memory 516. Alternatively, the classifier 532 may be pre-trained before being provided to the programme storage portion 518.

[0159] The second classifier 536 identifies which arsenic atoms 9, or ensembles of arsenic atoms 15, may be useful as entangled qubits. As a pre-processing step, the second classifier 536 determines the spacing between arsenic atoms 9 identified on the surface 5 of the substrate 1 by the first classifier 532. The second classifier 536 then classifies these based on their suitability for forming devices.

[0160] The second classifier 536 may adopt a binary classification (useful / not useful) or may adopt a grading of usefulness based on how closely the spacings match optimal or desired ranges, as discussed above. Depending on the application the threshold for which arsenic atoms 9 are used may be varied.

[0161] The output of the second classifier 536 includes an identification of which arsenic atoms 9, or ensembles or arsenic atoms 15, may be useful as entangled qubits. Also included in the output is an identification of modifications to be made to the surface distribution of defects to make the identified ensembles or arsenic atoms 15 useful and where to place electronic control structures on the same surface as the arsenic atoms 9.

[0162] The outputs of the second classifier 536 are used to control the STM 504 and hydrogen resist lithography process in the next steps 110, 112 of the method 100. For example, the output is used to automate control of deposition of the passivation layer, movement of the STM tip to define windows in the passivation layer, and exposure to precursor gas or other material deposition.

[0163] The output of the second classifier 536 may be provided in a datafile 538 stored in the data storage 520 portion of the memory 516 or in any other suitable form. In some examples, the output of the second classifier 536 may simply be provided to the processor 522 for controlling the surface modification 110, without being stored.

[0164] The training data for the second classifier 536 may comprise a database of spacings that may be suitable, with appropriate user applied labels for each set of spacings. Alternatively, the set of spacings may be characterised by a single parameter or feature extracted from the images. The training data may have the feature manually labelled by a user, and the second classifier 536 may be arranged to extract the same feature from captured images. In a third example, the training data may comprise a set of images of arsenic dopant atoms 9 with different spacing labelled appropriately. In this case, the classifier may work based on image matching. As with the first classifier, 532, training data for the second classifier may be stored as a datafile 536a in the data storage 518 portion of the memory 516. Alternatively, the classifier 536 may be pre-trained before being provided to the programme storage portion.

[0165] The third classifier 540 assesses the viability of arsenic atoms 9, 17 or ensembles of arsenic atoms 15 as qubits. The third classifier 540 may operate in any of the ways discussed above in relation to the first classifier 532 or the second classifier 536, and the training data for the third classifier 540 may be of similar form to first classifier 532 (i.e. manually cropped and labelled AFM images of samples that have undergone similar processes) the second classifier 536 (i.e. a database of spacings).

[0166] In addition to spacings and / or image analysis, the third classifier 540 may make use of electrical measurements captured in the AFM 506 using the electrical components laid down. The training data may include examples of suitable electrical characteristics.

[0167] It will be appreciated that whilst the second classifier assesses potential devices after deposition of the arsenic atoms 9, the third classifier makes the assessment after modification of the surface distribution of defects 13, and application of certain electrical structures and overgrowth of silicon.

[0168] The output of the third classifier 540 includes identification of which arsenic atoms 9, or ensembles of arsenic atoms 15, may be useful as qubits. This may be provided to a further system (not shown) for design of surrounding electrical components. This may be automated or by a user. Alternatively, the output of the third classifier 540 may include the design for the electrical components.

[0169] The output of the third classifier 540 may be provided in a datafile 542 stored in the data storage 520 portion of the memory 516 or in any other suitable form. In some examples, the output of the third classifier 540 may simply be provided to the next stage for design or formation of the electrical components.

[0170] In the above examples, the second and / or third classifiers 536, 540 may include designs for electrical contacts and the like in their outputs. It will be appreciated that the design of these components may be within the classifiers 536, 540 or in a separate module 540. In some cases, the function of the processing circuity 522 and / or memory 516 may be distributed over a number of connected units. These units may be connected over any suitable network connection, cloud based, as well as through the bus 526.

[0171] In the examples discussed above, a random surface distribution of arsenic atoms 9 is generated by exposing the clean surface 5 of the substrate 1 to arsenic precursor gas 11. A passivation player is then applied to create an atomic resolution image of the surface 5 to identify incorporation sites.

[0172] Figure 6 illustrates an alternative method 600 for forming atomic defect-based qubits using a random surface distribution of arsenic atoms 9.

[0173] According to the method 600 of Figure 6, a hydrogen passivation layer is formed in a first step 602.

[0174] The ratio of semiconductor atoms 7 in the surface layer to hydrogen atoms in any passivation layer is 1 : 1 since each silicon atom 7 on the surface 5 has a single dangling bond (not shown) which is terminated by a single hydrogen atom.

[0175] In a second step 604, the passivation layer is heated in a thermal annealing process. The heating diffuses and desorbs some of the hydrogen in the passivation layer to generate a random distribution of holes in the resist layer, the holes forming incorporation sites for arsenic. The density of incorporation sites can be controlled or modified by varying the time and temperature of the anneal.

[0176] The temperature of the anneal is above the thermal activation temperature to excite hydrogen vacancy diffusion and evaporation in the passivation layer. For example, the temperature may be between 320°C and 450°C. The anneal may be for any suitable duration, for example, one minutes.

[0177] The holes generated by the anneal are typically one or two dangling bonds in size, and the distribution of hole size may optionally be controlled by temperature variation. In a next step 606, the substrate 1 is exposed to precursor gas 11 to adhere arsenic atoms / moieties at the incorporation sites.

[0178] In a further step 608, an STM image is captured to obtain a spatial map of features on the surface 5 of the substrate 1, in a similar manner to the method 100 of Figure 1. In this example, the AsHxmoieties are mapped rather than the As-Si heterodimers, making the incorporation sites easier to discern from the passivation layer. As discussed above, each adsorbed moiety will be converted to an As-Si heterodimer on annealing so the presence of a moiety can be used to infer the position of an arsenic atom 9.

[0179] In the method of Figure 6, the precursor gas 11 the pressure and exposure time of the precursor gas 11 may be above or below saturation doses. Where the pressure and exposure time of the precursor gas 11 is above saturation doses, all holes in the resist are either filled by an AsHxmoiety (which is incorporated with high incorporation rate) or re-terminated by a hydrogen atom dissociated from the precursor gas. Thus, there is no need to expose to resist to terminate any dangling silicon bonds at sites where arsenic is not incorporated.

[0180] After the image is captured, the surface is annealed in step 610 to incorporate the arsenic atoms 9 without carrying out any modification. The method 600 then proceeds in the same manner as the method 100 of Figure 1, from the step of analysing to map the location of the arsenic atoms 9.

[0181] In the examples discussed above, a random surface distribution of arsenic atoms 9 is generated over the entire surface 5 of the substrate 1. However, this is by way of example only. In some embodiments, the locations in which the arsenic atoms 9 are incorporated into the surface may be defined using a resist process similar to the method 200 discussed in relation to Figure 3 or Figure 4.

[0182] For example, when forming atomic defect-based qubits based on the methods 100, 600 of Figure 1 or Figure 6, the resist process may define a specified area or areas (windows) over which the random distribution is provided. When using the method 100 of Figure 1, the surface 5 of the substrate 1 may be passivated with a resist layer and then one or more windows opened for forming the random distribution of arsenic atoms 9.

[0183] The windows may be large enough that the surface 5 of the lattice 1 behaves as a bulk surface. For example, the area of the window may be large enough that one or more arsenic atoms can be incorporated without the dissociation reaction being interrupted part way through leaving a AsHxmoiety rather than the arsenic atom.

[0184] When using the method 600 of Figure 6, a double layer resist may be used. A mask layer may be provided over the first resist layer (which forms the holes in the resist layer following the thermal anneal). The mask layer may prevent holes being formed in some regions of the first resist layer.

[0185] It will be appreciated that the band structure of silicon varies along at least one crystal direction of the lattice. Therefore, optionally, the spacing between defined windows for positioning of arsenic atoms 9 may be set such that the arsenic atoms 9 are always at positions with similar band structure, or the band structure for each arsenic atom 9 may be controlled in some other way by selecting of the spacing.

[0186] Alternatively, the resist / mask layer may simply be used to mask areas where arsenic is not desired. For example, the resist / mask layer may prevent defects being formed at the edges of devices, or where other electrical components are to be provided.

[0187] In further examples, when depositing arsenic atoms 9 in windows defined in a resist layer, the windows may be smaller than the size required for the surface 5 to behave as a bulk surface. The sites may be of any desired size below this threshold. For example, the windows may be two silicon atoms 9 in size, to allow precise placement of only single atoms, instead of generating a random distribution.

[0188] As discussed above, the incorporation of arsenic atoms 9 at sites of this size is approximately 75%. In the methods discussed above (and in WO 2023 / 175315) the process of defining windows and exposing to precursor gas 11 may be repeated to achieve greater incorporation. However, in the current embodiment, the deposition of arsenic in 75% of sites may simply be used to generate an arbitrary distribution that can be mapped at later stages of the method.

[0189] As the size of the sites for incorporation increases, the rate of incorporation increases, but the accuracy of placement reduces (and the chances of incorporating multiple atoms increases) until the window reaches bulk size and a random distribution forms.

[0190] Where arsenic atoms 9 are provided at precisely defined sites, modification of the distribution of defects 13 may also include removal of adsorption site in the resists (one or two exposed silicon atom sites) by selective re-termination of the surface at these sites, prior to exposure to precursor gas 11.

[0191] In this case, an atomic resolution image (for example an STM) is captured before the exposure to the precursor gas. To achieve this, the first classifier 532 maps adsorption sites in the resist where substitution defects can be introduced. Adsorption sites (resist vacancies) may be a further class of defect identified by the first classifier 532, in addition to silicon atoms 7, arsenic dopant atoms 9, other types of dopant / impurity, different types of defects and the like.

[0192] Selective re -termination is achieved using, for example an STM or AFM tip. Various techniques for selective re-termination are known such as, for example, the technique disclosed in “Atomic White-Out: Enabling Atomic Circuitry through Mechanically Induced Bonding of Single Hydrogen Atoms to a Silicon Surface", TR Huff et al, ACS Nano 2107, 11, 9, 8636-8642, which is incorporated by reference.

[0193] Selective re -termination is performed prior to adsorption of arsine and incorporation of arsenic atoms 9.

[0194] Following selective re-termination, the step of obtaining an atomic resolution image after exposing to the precursor gas may be omitted or may still be performed. In this case, the step of capturing the image may locate AsHxmoieties and / or As-Si heterodimers.

[0195] It will be appreciated that the same classifier 532 may be used to identify resist vacancies as other defects, or different classifiers may be used. As discussed, the classifier may identify multiple classes, or may operate on a binary classification (absorption site or no absorption site).

[0196] By knowing the placement of arsenic atoms 9 with some precision (either in large bulk areas or precisely defined windows), the imaging stages can be sped up, by only imaging in the areas that have been exposed to precursor gas 11.

[0197] In the examples discussed above, the imaging steps are performed prior to any anneal to incorporate the arsenic atoms. Therefore, where bulk regions of the surface are exposed to precursor gas, As-Si heterodimers are imaged, and where smaller regions are exposed, AsHxmoieties are imaged.

[0198] This is by way of example only. Imaging may occur after annealing to incorporate adsorbed arsenic. When imaging is performed after an anneal, As-Si heterodimers will be detected whether the exposed regions of the surface are bulk sized or smaller. Furthermore, additional features which may be indicative of incorporated arsenic atoms may be detected. For example, ejected silicon atoms and dangling bonds in the surface may be detected.

[0199] As discussed above, training data for any classifier should be captured under similar conditions, and so any classifier for analysing images after an anneal step should be trained in training data including images captured after an anneal, and appropriately labelled.

[0200] In the examples discussed above, the surface 5 of the substrate 1 is heated to incorporate the arsenic atoms. It will be appreciated that this may be omitted, and the atoms may be incorporated by the anneal after modification of the surface.

[0201] As will be appreciated by the person skilled in the art any surface 5 of a silicon lattice may also include naturally occurring defects. In at least some embodiments, the naturally occurring defects may also be included as target defects for forming qubits, if the defect is of the correct sort (as determined by the first classifier 532). Naturally occurring defects may also be incorporated in other electronic structures. In the above method, the electronic control structures are formed after deposition of the arsenic atoms 9. However, at least some of the electronic control structures may be deposited prior to forming the defect distribution. In this case, the step 108 of classifying target / arsenic defects 13 based on their suitability to be used for qubits may also include the spacing from the pre-deposited control structures as a factor in the assessment.

[0202] In the above method 100, STM imaging is used to provide a spatial map of surface features and AFM imaging is used to provide a spatial map of wavefunction overlap and optionally charge trap defects. Both maps are used at different stages to assess the viability of defects 13 or groups of defects 13 as potential devices. It will be appreciated that the use of these technologies is by way of example only. Any suitable atomic resolution imaging technique may be used in one or both of these steps. The steps may use different imaging techniques or the same imaging techniques.

[0203] Where different imaging techniques are used to capture images that are provided to a classifier, the classifier should be trained on images / data captured using the same technique, to ensure images are captured in the same regime.

[0204] The maps of defects (spatial map of surface features and / or spatial map of wavefunction overlap) may occur at various different stages.

[0205] For example, the spatial map of surface features may be obtained before and / or after the arsenic atoms 9 are incorporated into the surface 5 of the substrate 1. When imaging before incorporation, the moieties of the precursor gas are imaged. When imaging after incorporation, incorporation reaction products are imaged (such as arsenic -silicon heterodimers or ad-Si atoms or clusters where silicon is ejected from the surface.

[0206] The wavefunction map may be obtained before or after silicon overgrowth.

[0207] In the example discussed above, the method includes the step 110 of modifying the surfacing distribution of defects. It will be appreciated that this step may be omitted, such that qubits may be made using the distribution of arsenic atoms 9 from the first deposition step 102. In this case, the surface is annealed to incorporate the arsenic atoms 9 without carrying out any modification. In the example discussed above, various processes for modifying the surface distribution are discussed including a hydrogen resist photolithography process, and removal of atoms by an STM tip. It will also be appreciated that this is by way of example only. The surface distribution may be modified by any suitable technique. This may include, for example, addition of atoms using any suitable photolithography techniques, removal of atoms by any suitable etching (with a mask defined by photolithography), and other techniques for atomic scale manipulation.

[0208] The controller 502 uses classifiers for identifying and mapping defects, assessing the potential of defects as qubits, and assessing the viability of ensembles as qubits. The use of classifiers is by way of example only. Any suitable machine learning algorithm may be used instead.

[0209] In further embodiments, at least some of the identification, mapping and assessment need not use machine learning at all. Various pattern matching techniques may also be used. Furthermore, where parameters such as spacings are measured, these may simply be checked against desirable ranges or thresholds.

[0210] In the example discussed above, the method is used to make qubits, which can be used for data processing or storage. The method 100 discussed above can be scaled to provide any number of arsenic atoms 9 in a surface. For example, a silicon wafer (not shown) having diameter between 1 inch / 25.4mm to 12 inch / 300mm could be provided with an array of arsenic atoms 9 and cleaved into separate chips or devices. By this method, a number of chips having hundreds or thousands of qubits each may be made.

[0211] In the above, single atoms are incorporated for providing qubits. However, it will be appreciated that qubit (or devices for control or readout of qubits) can also be formed by quantum dots or wires or two-dimensional electron gasses (sheets), formed by small clusters of arsenic atoms 9 which still exhibit quantum properties. The above method 100 may also be used to form randomly distributed qubits in this way.

[0212] Devices other than qubits may also be formed. In other examples, qudits (which have more than two states) may be formed using the 3 / 2 nuclear spin of arsenic atoms 9. Other quantum information processing devices may also be formed. In the above examples, the qubits or other devices are based on electron or nuclear spin. However, it will be appreciated that the arsenic atoms may also make quantum devices based on other properties.

[0213] In the above examples, arsenic is provided as the dopant atom. However, this is by way of example only. In other cases, other suitable dopants may be used. Suitable dopants include, but are not limited to, group III and group V elements.

[0214] For example, phosphorus or boron may be used, with posphine or diborane used as the dopant precursor. For example, phosphorus or boron = may be used when exposing the entire surface 5 of the substrate 1 (or windows large enough to function as a bulk surface) to incorporate dopant.

[0215] In the examples discussed above, a single species of dopant is used (e.g., phosphorus or arsenic). In other examples, there may be two or more species of dopants.

[0216] In the above, the dopant atoms 9 are incorporated into the (001) silicon surface. It will be appreciated that dopant atoms 9 may be incorporated into other surfaces on silicon.

[0217] It will also be appreciated that arsenic dopants may also be incorporated into other semiconductor lattices 1. For example, a germanium lattice may be used instead of silicon. Any group IV semiconductor lattice may be used, and other compound or atomic semiconductors.

[0218] In some of the above methods, a step of thermal annealing is used to incorporate the adsorbed dopant atoms 9. However, this is by way of example only. Any suitable method of energy transfer may be used. For example, energy may be transferred through an STM tip or an electron beam.

[0219] In the examples discussed above, the passivation layer / resist is hydrogen and the molecules of the precursor gas include hydrogen in addition to phosphorus or arsenic. On other examples, different passivation layers may be used, and the molecules of the precursor gas may include the same or different atoms as the passivation layer. Various computation and modelling packages may be used to help identify clusters of two or more defects to form qubits, locations in which the defect distribution may be modified to form qubits, and design electronic control structures and other parts of the device. For example, QTCAD from Nanoacademic Technologies Inc may be used.

[0220] In the examples discussed above, STM and AFM tips are used for atomic scale imaging and manipulation. Processes for automatic alignment and operation of the STM tip will be apparent to the person skilled in the art. In one example, an array of cantilevered STM tips or AFM tips may be used for faster processing. See, for example, P. Vettiger et al., "The “Millipede” — More than thousand tips for future AFM storage," in IBM

[0221] Journal of Research and Development, vol. 44, no. 3, pp. 323-340, May 2000, doi: 10. 1147 / rd.443.0323 or R. Huff et al., “Atomic White-Out: Enabling Atomic Circuitry through Mechanically Induced Bonding of Single Hydrogen Atoms to a Silicon Surface” in ACS Nano, vol. 11, no. 9, pp. 8636-8642, July 2017.

Claims

Claims1. A method of manufacturing a plurality of quantum information processing devices on a substrate, the method including: acquiring an atomic scale spatial map of the substrate surface; using one or more electronic processors, processing the acquired atomic scale spatial map to identify defects in the substrate surface, establishing which of the identified defects are target defects and determining the location of the target defects; using the one or more electronic processors to classify the target defects based on their suitability for forming quantum information processing devices, the one or more electronic processors forming a classifier, the classifier having been pre-trained with a training data characterising quantum information processing devices; and defining electrical components to control and readout quantum information processing devices formed by at least some of the defects classified as suitable for forming quantum information processing devices.

2. The method of claim 1, comprising: prior to acquiring the atomic scale spatial map, incorporating dopant atoms into the substrate surface to form the target defects.

3. The method of claim 2, wherein incorporating dopant atoms comprises: exposing the substrate surface to a dopant precursor gas to form a random distribution of target defects over the substrate surface.

4. The method of claim 3, comprising: controlling the supply of dopant precursor gas to control the surface density of target defects base and average spacing between target defects.

5. The method of any of claims 2 to 4, wherein incorporating dopant atoms comprises: forming a passivation layer on the surface of the substrate; andheating the passivation layer, to generate a random distribution of holes in the passivation layer, the holes forming incorporation sites for dopant atoms.

6. The method of claim 2, wherein incorporating dopant atoms comprises: exposing predefined regions of the substrate surface to a dopant precursor gas to form a quasi-random distribution of target defects over the substrate surface, optionally wherein the predefined regions are formed by masking areas that are not to be exposed to the dopant precursor gas.

7. The method of claim 5 or claim 6, wherein each of the predefined regions is arranged to incorporate single atoms only.

8. The method of claim 6 or claim 7, wherein: the predefined regions are arranged in an array with regular spacing along at least one direction along the substrate surface.

9. The method of claim 8, wherein: the substrate comprises a semiconducting material; the band structure of the semiconducting material varies along at least one crystal direction of the substrate; and the spacing between predefined regions corresponds to the variance of the band structure along the at least one crystal direction of the substrate.

10. The method of any preceding claim, further including: after classifying the target defects, modifying the distribution of defects on the substrate surface.

11. The method of claim 10, wherein modifying the distribution of defects on the surface comprises: creating additional target defects at defined locations.

12. The method of claim 11, wherein creating additional defects comprises placing dopant atoms in an atomic surface layer of the substrate.

13. The method of claim 11 or claim 12, wherein the additional target defects are created using a scanning tunnelling microscope.

14. The method of claim 13, wherein the scanning tunnelling microscope is also used to acquire the atomic scale spatial map of the substrate surface, and the substrate is kept under vacuum in a chamber of the scanning tunnelling microscope between acquiring the atomic scale spatial map and generating target defects at defined locations.

15. The method of any of claims 10 to 14, wherein classifying the target defects based on their suitability for forming quantum information processing devices includes: classifying targets based on their suitability using the identified target defects; and classifying targets based on their suitability if the distribution of defects on the substrate surface is modified.

16. The method of any preceding claim, wherein target defects comprise a controllable quantum property to function as a quantum information processing device or part of a quantum information processing device, preferably wherein the controllable quantum property comprises a property selected from: spin or charge or energy.

17. The method of claim 16, wherein a target defect comprises a dopant atom.

18. The method of: claim 2 or any claim dependent thereon; or claim 12 or any claim dependent thereon; or claim 17; wherein the dopant comprises arsenic or phosphorus atoms.

19. The method of any preceding claim wherein target defects are established using the one or more electronic processors to classify the types of defect, the one or more electronic processors forming a second classifier, the second classifierhaving been pre-trained with a training set of images of labelled defects and images without defects.

20. The method of any preceding claim, wherein the classification of the suitability of target defects for forming quantum information processing devices is based on the proximity of a defect to neighbouring defects, such that interactions between the defects are suitable for quantum data processing.

21. The method of claim 20 wherein the classification of the suitability of target defects for forming quantum information processing devices is based on the proximity of an ensemble of two or more defects, such that interactions between the two or more defects are suitable for quantum data processing.

22. The method of any preceding claim including: overgrowing the substrate surface to encapsulate the quantum information processing devices formed by target defects classified as being suitable for forming quantum information processing devices.

23. The method of claim 22, further including: after overgrowing the substrate surface; capturing a spatial map of wavefunction overlap of the overgrown surface; and assessing the viability of quantum devices that will be formed by target defects classified as being suitable for forming quantum information processing using a third classifier, the third classifier having been pretrained with a training set of labelled images and / or data characterising quantum information processing devices, the method further optionally including: defining electrical components to control and readout quantum information processing devices based on the assessed viability.

24. The method of any preceding claim, wherein at least some of the electrical components to control and readout quantum information processing devices are formed in the substrate surface; and / orat least some of the electrical components to control and readout quantum information processing devices are formed in layers above or below the substrate surface.

25. A non-transient computer-readable medium containing instructions which, when run on a system including processing circuitry, cause that system to implement the classifier of any preceding claim.

Images