WO2014049414A1

WO2014049414A1 - Direct bonding process

Info

Publication number: WO2014049414A1
Application number: PCT/IB2013/002089
Authority: WO
Inventors: Marcel Broekaart; Arnaud Castex; Hamza LASRY
Original assignee: Soitec
Priority date: 2012-09-27
Filing date: 2013-09-20
Publication date: 2014-04-03
Also published as: CN104662649B; CN104662649A; FR2996052B1; FR2996052A1

Abstract

The invention relates to a direct bonding process, which comprises: positioning a first wafer (10) on the surface of a chuck (2), said surface comprising grooves (4); applying in the grooves (4) a first pressure lower than a second pressure seen by the exposed side of the first wafer (10); and bringing a second wafer (16) into contact with the exposed side of the first wafer (10), then initiating the propagation of a bonding wave between the two wafers while the first and second pressures are maintained.

Description

Direct bonding process

Background of the invention

The present invention relates to the fabrication of multilayer structures formed by joining to a first wafer (carrier substrate or wafer) at least one second wafer (or substrate) by direct bonding (also known as molecular bonding). Such heterostructures are especially used in microelectronics or optoelectronics.

The invention more particularly relates to the direct bonding of semiconductor wafers in the context of the fabrication of multilayer structures, for example SOS (silicon-on-sapphire (AI2O3)) or GaNOS (GaN-on-sapphire) structures.

Such multilayer structures (also called multilayer semiconductor wafers) are typically produced using three-dimensional integration (3D-integration) technology involving the transfer to a first wafer, called the final substrate, of at least one layer formed from a second wafer, this layer corresponding to a portion of the second wafer in which elements, for example a plurality of microcomponents, have been formed, the first wafer either being a blank wafer or comprising other corresponding elements.

As is known, direct bonding to a first wafer (carrier wafer) of a second wafer may generate strains of various types in the multilayer structure obtained. Such bonding may, for example, generate nonuniform strains in the first and second wafers thereby making it more difficult to align components or other patterns formed in the second wafer with the underlying carrier wafer. The formation mechanism of this misalignment (or overlay) resulting from nonuniform strains is for example described in patent application FR 2 965 398.

Direct bonding may moreover generate bow in the joined wafers forming the multilayer structure. This bow may especially appear during a heat treatment because of differences in the thermal properties (different thermal expansion coefficient (TEC), etc.) of the various wafers joined in the multilayer structure.

Mention may be made, for example, of patent application FR 2 954 585, which especially describes the behaviour of a heterostructure during a bond strengthening anneal carried out at a temperature of about 160°C, the heterostructure being formed by bonding a first SOI (silicon-on-insulator) wafer to a sapphire substrate. The difference between the TEC of silicon, the main component of the SOI structure, and the TEC of sapphire leads to the assembly bowing during the heat treatment such that high debonding stresses are exerted at the edges of the heterostructure.

These stresses lead to unsatisfactory transfer at the edge of the wafer, possibly resulting in the formation of an excessively large and irregular "crown" (i.e. a zone in which the first wafer has not transferred to the carrier substrate) which may especially lead to flaking at the edge of the wafers.

Patent application FR 1 153 349 moreover describes the case where a layer from a first substrate (called the "substrat donneur") is transferred to a second substrate (called the "substrat accepteur"), the second substrate having previously undergone various technological processes (formation of cavities, etc.). To transfer this layer the first and second substrates are direct bonded, the multilayer structure thus obtained is annealed, and the structure is then subjected to a chemical-mechanical thinning step. The bow in the final multilayer structure is mainly a result of the initial strain in the second substrate (i.e. the "substrat accepteur"), this initial strain resulting from the technological processing steps (etching, deposition, etc.) undergone by the second substrate before bonding.

However, this bow effect observed in multilayer structures has substantial drawbacks. The stresses generated may in particular lead to (partial or complete) debonding or cracking of the multilayer structure during technological processing steps (heat treatments, etc.) carried out after the direct bonding of the first and second wafers. The parameters of post-bonding technological processes (heat treatments, thinning etc.) carried out on the multilayer structure must therefore be chosen with caution in order to avoid excess stress, thereby substantially increasing the complexity of these processes and how difficult they are to control/and therefore how much they cost.

When, for example, the carrier wafer comprises cavities, fabs generally require the multilayer structure post-bonding (the first wafer bonded to the carrier wafer being free of any component) to respect certain bow criteria after thinning, in order to allow subsequent technological processes to be carried out under acceptable conditions..

At the present time there is no satisfactory way to control, at least to a certain extent, the direction and amplitude of the bow in the wafers of a multilayer structure bonded by direct bonding. Summary of the invention

For this purpose, the present invention relates to a direct bonding process, which comprises:

- positioning a first wafer on the surface of a chuck, the surface comprising grooves;

- applying in the grooves a first pressure lower than a second pressure seen by the exposed side of the first wafer; and

- bringing a second wafer into contact with the exposed side of the first wafer, then initiating the propagation of a bonding wave between the two Wafers while the first and second pressures are maintained.

The pressure difference ΔΡ thus applied induces local stresses in the first wafer which deforms locally in particular at the bonding surface of the first wafer (i.e. the surface to be bonded with the second wafer). During propagation of the bonding wave, the second wafer conforms to the curvature imposed by the first wafer and is in turn subjected to the local stresses induced by the pressure difference ΔΡ.

Applying these local stresses to the bonding, interface advantageously allows the bow in the multilayer structure to be controlled, to a certain extent, after bonding. As will be explained in more detail below, the bow in the multilayer structure obtained is systematically concave with respect to the reference plane formed by the surface of the chuck.

The final bow in the final structure is therefore less dependent on the specific curvatures (i.e. before bonding) of the first and second wafers. Greater bow uniformity may thus be obtained over a plurality of multilayer structures, in the same batch for example.

The ability to control the direction of the curvature of the multilayer structure (and to a certain extent the amplitude of this curvature) makes it possible to respect the continually increasing requirements of fabs and to prevent defective bonding or debonding, especially in the case where technological steps (heat treatments, etc.) are subsequently carried out on the multilayer structure.

As a first variant, the grooves are arranged in the form of an orthogonal grid over the entire surface of the chuck.

As a second variant, the grooves are arranged in the form of concentric annular grooves centred on the centre of the chuck. In one particular embodiment, the grooves are uniformly distributed over the entire surface of the chuck. Such a distribution allows uniform stresses to be applied to the bonding interface between the first and second wafers.

In another embodiment, the grooves are more closely spaced in one zone of the surface of the chuck than the grooves in the rest of the surface of the chuck. In this way, the curvature of the multilayer structure is increased locally in the zones of the chuck where the grooves are more closely spaced.

This zone where the grooves are more closely spaced may for example correspond to a ring on the periphery of the chuck, in order to increase the curvature of the multilayer structure at the edge of the wafer.

Preferably, the pressure difference between said first and second pressures is greater than or equal to 3 mbar. In one particular embodiment, this pressure difference is between 3 and 10 mbar.

In another particular embodiment, the chuck heats the first wafer at least during the contacting step and the step of initiating the propagation of the bonding wave. This heating may also be carried out during the step in which the pressure difference is applied.

The bonding process according to the invention may furthermore comprise:

- annealing the multilayer structure resulting from the direct bonding of the first and second wafers; and

- thinning the first wafer or the second wafer.

Brief description of the drawings

Other features and advantages of the present invention will become apparent from the following description given with reference to the appended drawings which illustrate a nonlimiting embodiment. In these figures:

- Figures 1A to IF are cross-sectional views schematically showing each step (S10-S30) of a bonding process according to one particular embodiment of the invention;

- Figure 2 shows, in the_, form of a flowchart, the main steps of the embodiment illustrated in Figures 1A-1F;

- Figures 3A and 3B show cross-sectional views of two example wafers exhibiting concave and convex bow, respectively; and - Figure 4 is a graph showing, in the form of a curve, the variation in the bow in a multilayer structure as a function of the pressure difference ΔΡ applied according to one particular embodiment of the invention. Detailed description of embodiments of the invention

The present invention relates to the fabrication of multilayer structures by direct bonding of a first wafer (or a carrier wafer) with a second wafer. The invention especially applies to the formation of SOS or GaNOS multilayer structures, for example.

At least one of the wafers forming the multilayer structure may comprise at least one microcomponent having been produced before the bonding. For the sake of simplicity, the term "microcomponents" is, in the rest of this text, understood to mean the devices or any other patterns resulting from technological steps carried out on or in the layers, and the position of which must be controlled with precision. They may therefore be active or passive components, simple patterns, contact pads, interconnects, or even microchannels or cavities.

In order to better control the bow in a multilayer structure formed by direct bonding, the present invention proposes to apply local stresses to the bonding interface.

By studying the formation mechanism of the strains generated in multilayer structures, as described above/ the Applicant has indeed observed that applying certain stresses to the bonding interface during the direct bonding operation allows the bow in the resulting multilayer structure to be controlled to a certain extent. Controlling this bow advantageously allows the general shape of the wafers forming the multilayer structure to be corrected, and the shape of the structure to be anticipated even before it is direct bonded.

Thus, the Applicant has developed a direct bonding process allowing such stresses tq be applied in a way that allows bow in multilayer structures to be controlled. As explained in greater detail below, this process in particular involves a chuck comprising grooves on its contact surface (i.e. the surface of the chuck intended to make contact with the first wafer of the multilayer structure to be produced).

One particular embodiment of the invention is now described with reference to Figures 1A to IF and 2.

Figure 1A shows a chuck 2 comprising grooves 4, here distributed uniformly over the entire contact surface 6 of the chuck 2. In this example, the grooves 4 take the form of an orthogonal grid made up of two sets 4A and 4B of parallel grooves uniformly distributed over the entire surface 6, these two sets of grooves lying perpendicular to each other. As described in greater detail below, variants can be envisaged in the context of the invention as regards the distribution of the grooves and/or the dimensions and shapes of these grooves.

In the example considered here, the grooves 4 are each 5 mm in width and 1 mm in depth. However, it will be understood that grooves with other dimensions may be envisaged in the context of the invention.

The grooves 4 are here equipped with suction means 8 which will be described in greater detail below.

Figure IB shows a first 150 mm diameter wafer 10 (or carrier wafer) positioned on the surface 6 of the chuck 2 (S10). Other wafer diameters (200 mm, 300 mm, etc.) or shapes may naturally be envisaged.

In this example, the wafer 10 is an SOI (silicon-on-insulator) wafer and comprises a silicon layer on a carrier that is also made of silicon, a buried oxide layer (for example made of SiO₂) being placed between the silicon layer and the silicon carrier. However, it will be understood that the first wafer 10 may consist of a multilayer structure of another type or of a monolayer structure.

Moreover, the carrier wafer 10 here has a specific curvature Kl i.e. an initial curvature before bonding.

Specifically, it will be recalled that before bonding each wafer has a specific curvature which may be concave, as for the wafer 100 in Figure 3A, or convex, as for the wafer 110 in Figure 3B. This curvature defines the bow in the wafers. The bow may for example be paraboloidal (and especially spherical) in shape. . "

As illustrated in Figures 3A and 3B, the bow Δζ in a wafer corresponds to the distance (arrow) between a (typically perfectly flat) reference plane P on which the wafer rests freely, and the wafer itself. With the wafer diameters conventionally used in the semiconductor field, namely between a few tens of millimetres and 300 millimetres, bow is measured in microns (pm) whereas curvature is generally measured in m^"1 or km^"1 because the curvature of the wafers used in the semiconductor field is very small and therefore the corresponding radius of curvature is very large.

In the example in Figure IB, the bow Kl in the carrier wafer 10 is concave relative to the surface 6 of the chuck 2 (Kl < 0). Once the first wafer 10 has been positioned on the chuck 2, a first pressure PI is generated (SI 5) in the grooves 4 using suction means 8 (Figure 1C). This pressure PI is thus applied locally over the surface 10a of the carrier wafer 10 level with each groove 4. The suction effect is obtained here by pumping out the air 12 present in the grooves between the wafer 10 and the chuck 2, this air 12 being removed through orifices belonging to the suction system 8 and housed in the bottom of the grooves 4 in the chuck 2.

Alternatively, any other suitable means allowing the pressure PI to be applied locally level with the grooves could be used.

According to the invention, the applied first pressure PI must be such that PI is smaller than P2, where P2 is the pressure seen by the exposed side 10b of the first wafer 10. In the present case, P2 corresponds to the pressure in the chamber in which the bonding process of the invention is carried out.

The pressure difference ΔΡ=Ρ2 - PI applied (S15) locally to the first wafer 10 level with each groove 4 induces local stresses (or actions) 14 in the first wafer 10. Under the effect of these stresses, the first wafer 10, and in particular its exposed surface 10b, then deforms locally as schematically shown in Figure 1C. An enlargement (magnified portion) in Figure 1C shows one of the zones of the first wafer 10 level with a groove 4 and the stresses 14 induced locally.

It will be understood that here the stresses 14 translate physically into a force that locally presses the wafer 10 against the chuck 2, thereby producing slight strains in the wafer 10 mainly in the region of the grooves (slight deflection of the wafer 10 toward the bottom of the grooves). These slight strains generate a bow in the entire wafer 10, in particular in its exposed surface 10b, this bow depending on the physical arrangement of the grooves (width, orientation, distribution of the grooves over the surface 6, number of grooves, etc.).

In this example, the difference ΔΡ applied locally to the wafer 10 is preferably greater than or equal to APmin=3 mbar, and even more preferably between 3 and 10 mbar, inclusive. By applying a pressure difference ΔΡ greater than or equal to APmin, the wafer 10 is guaranteed to be securely fastened against the chuck 2 (clamping effect).

It will be noted that it can be envisaged to obtain the desired ΔΡ in a number of ways. In one particular embodiment, a pressure PI is applied by sucking the air from under the wafer 10, then the pressure P2 in the chamber is reduced until the desired APIs obtained. Once the desired pressure difference ΔΡ has been applied locally to the first wafer 10 (S15), the pressures PI and P2 are kept constant and the process continues with the direct bonding (S20) of a second wafer 16 to the deformed first wafer 10 (Figure ID). The same pressure difference ΔΡ as that applied in the preceding step S15 is therefore maintained during the bonding step S20 in the regions of the wafer 10 corresponding to the grooves 4.

No stress is exerted on the second wafer 16 before it is bonded to the first wafer 10. The wafer 16 is simply placed on the first wafer 10, in order to achieve the direct bonding, with no prior strain.

Direct bonding is a technique that is well known per se. It will be recalled that the principle of direct bonding is based upon bringing two surfaces into direct contact, i.e. no intermediate material (adhesive, wax, braze, etc.) is used. Such an operation requires the surfaces to be bonded to be sufficiently smooth, free from particles or contamination, and for them to be brought close enough together to allow contact to be initiated - typically a distance smaller than a few nanometres is required. Under these circumstances, the attractive forces between the two surfaces are strong enough to cause a bonding wave to propagate, the propagation of this wave leading to direct bonding (this bonding is due to attractive forces (Van der Waals forces) generated by electronic interactions between atoms or molecules in the two surfaces to be bonded).

Therefore, during this step S20 the second wafer 16 is brought into contact with the surface 10b of the first wafer 10, then propagation of a bonding wave is initiated at the interface between the wafers 10 and 16. The wafers are brought into contact and the wave propagation initiated while the same ΔΡ as that applied locally in step S15 is maintained. The technique used for bonding wave initiation is well known perse and will not be described in more detail here.

The second wafer 16 is, in this example, made of sapphire, and also has a diameter of 150 mm. However, the second wafer could consist of a monolayer structure of another type, or of a multilayer structure. As shown in Figure ID, the second wafer 16 has a convex specific bow K2 before bonding (K2 > 0). However, it may be envisaged, for example, for the bow K2 to be concave or for the wafer to be approximately flat.

Once propagation of the bonding wave has been initiated, the second wafer 16 conforms to the curvature imposed by the first wafer 10 during progression of the bonding wave (Figure ID). Once the bonding has terminated, a multilayer (or stacked) structure 20 of the SOS type is obtained containing the first wafer 10 and the second wafer 16, this structure having the desired bow KF.

The amplitude of the bow KF obtained is directly proportional to the local strains generated in the wafers 10 and 16 during the bonding process of the invention. According to the invention, whatever the form of the specific bow (concave, flat or convex) in the first and second wafers 10 and 16 before bonding, a multilayer structure 20 exhibiting a concave bow KF is obtained at the end of the bonding operation S20.

Furthermore, the greater the pressure difference ΔΡ, the more the bow

KF in the resulting multilayer structure 20 will be pronounced. Figure 4 shows the variation in the bow as a function of the value of ΔΡ applied in steps S15 and S20 in the embodiment envisaged here.

As indicated above, the pressure difference applied in steps S15 and S20 is chosen so that ΔΡ > ΔΡητιϊη. However, the value of ΔΡιτιίη used in particular depends on the thickness of the wafers 10 and 16 to be bonded and on the materials from which they are made. In the present case, the wafers 10 and 16 are made of silicon and are each 775 pm in thickness, and APmin is set to about 3 mbar. In this particular example, the concave bow in the multilayer structure 20 is between 38 pm and 85 pm, after bonding, for a pressure difference ΔΡ varying between 3 mbar and 900 mbar, respectively (see Figure 4).

In order to strengthen the bonding force between the two wafers 10 and 16, it is possible to then subject (S25) the multilayer structure 20 to a moderate heat treatment (at below 500°C, for example). In this example, an anneal for stabilising the bonding interface is carried out at a temperature between 140 and 150°C. This heat treatment allows the strength of the bonding between the wafers 10 and 16 to be increased and makes subsequent thinning of one or other of them possible under acceptable conditions. The bond strength may for example reach 400 mJ/m² after such an anneal..

As shown in Figure IF, the first wafer 10 is then thinned (S30) using a conventional method in order to obtain the wafer 11. In this example, the upper layer of the SOI first wafer 10 is removed by chemical-mechanical polishing (CMP), the buried insulating layer of the wafer 10 advantageously serving as a chemical etch stop layer in order to set the thickness of the remnant wafer 11. The final thickness of the wafer 11 may for example be between 4 and 10 pm. Alternatively, the wafer 10 may be thinned in another way such as by chemical etching or by cleaving along a weakened plane formed in the wafer 10 beforehand, for example by ion implantation (e.g. implantation of H or He impurities and cleaving according to the SmartCut® technology).

In the example in Figure IF it was the first wafer that was thinned.

However, as an alternative it may be envisaged to thin the second wafer in step S30.

Thus a three-dimensional SOS structure 20 is obtained formed from the second wafer (here the carrier substrate) and a layer 11 corresponding to the remnant portion of the first wafer 10.

Microcomponents (not shown) may then be formed in the transferred layer 11. These microcomponents are formed using conventional methods, typically by photolithography by means of at least one mask defining zones for forming patterns corresponding to all or part of the microcomponents to be produced. A tool, such as a stepper, providing selective irradiation is in general used to irradiate the zones or patterns intended to be produced.

In the end, applying local stresses to the first wafer, and more particularly to the bonding interface with the second wafer, advantageously allows the bow exhibited by the multilayer structure after bonding to be controlled to a certain extent. As indicated above, the bow in the multilayer structure is systematically concave (see the wafer 100 in Figure 3A) with respect to the reference plane formed by the contact surface 6 of the chuck 2. The final bow KF is therefore no longer as dependent on the specific curvatures Kl and K2 of the first and second wafers. Greater uniformity in the bow KF exhibited .by a plurality of multilayer structures in the same batch may thus be obtained. This in particular makes it easier to subsequently carry out technological steps on the multilayer structures thus produced.

The ability to control the direction of the curvature of the multilayer structure (and to a certain extent the amplitude of this curvature) makes it possible to respect the continually increasing requirements of fabs and to prevent defective bonding or eventual cracking, especially in the case where technological steps (heat treatments, etc.) are subsequently carried out on the multilayer structure.

In one variant, the chuck 2 (or holder) is configured to heat the carrier layer 10 in the contacting and bonding wave initiation step S20 (and optionally also in the preceding step S15 in which ΔΡ is applied). Applying heat by means of the chuck 2 allows the effect that generates the concave bow in the final multilayer structure 20 to be enhanced relative to the same process without heating. The chuck is preferably heated to a temperature between room temperature (20°C for example) and 200°C.

It will be understood that it is possible to vary the spatial configuration of the grooves in the surface of the chuck and to vary the difference ΔΡ applied in order to control to a certain extent the value KF of the final bow. It is in particular possible to adjust at least one of these parameters in order to tend toward the desired bow:

- the width of the grooves;

- the number of (or density of the) grooves arranged in the surface of the chuck;

- the orientation of the grooves;

- the distribution of the grooves over the entire contact surface of the chuck, etc. (as a variant, one or more grooves may be arranged in spirals, or the grooves they be arranged in a spider-web pattern). As indicated above, the orientation of the grooves may correspond to an orthogonal (or optionally non-orthogonal) grid (or chequerboard). Alternatively, the grooves may be arranged in concentric annular rings.

In addition, the grooves may advantageously be placed uniformly over the entire surface of the chuck in order to apply stresses that are as uniform as possible to the bonding interface. The grooves may for example be arranged in a uniform orthogonal (or non-orthogonal) grid or, alternatively, in the form of uniform concentric annular rings, so that the grooves are separated from each other by the same distance over the entire surface of the chuck.

However, it is possible to envisage variants in which the grooves are distributed nonuniformly over the surface of the chuck.

The grooves may for example be configured to be spaced more closely together in a particular zone of the surface of the chuck than the grooves in the rest of the surface of the chuck. This configuration makes it possible to increase locally the curvature in the multilayer structure in the zones of the support where the grooves are more closely spaced.

This zone in which the grooves are more closely spaced may correspond, for example, to a peripheral ring of the chuck in order to increase the curvature of the multilayer structure at the edge of the wafer. Other zones of the chuck may be envisaged depending on the specific circumstances of each situation.

Moreover, any type of trench or other similar indent may play the role the grooves play in the invention insofar as their dimensions allow the first wafer to be deformed locally by the application of the pressure difference ΔΡ described above. The dimensions chosen for the grooves may therefore also depend on the mechanical properties of the first wafer (and optionally also of the second wafer).

The grooves of the invention may, for example, be produced in the holder (chuck) by removing material from the surface of the latter by machining or any other technique. Alternatively, the grooves may be formed by adding material to the surface of the chuck or by forming protrusions on the surface in order to define the contours of the various grooves.

Claims

1. Direct bonding process, which comprises:

- positioning (S10) a first wafer (10) on the surface (6) of a chuck (2), said surface comprising grooves (4);

- applying, in the grooves a first pressure (PI) lower than a second pressure (P2) seen by the exposed side (10b) of the first wafer (10); and

- bringing a second wafer (16) into contact (S20) with the exposed side (10b) of the first wafer (10), then initiating the propagation of a bonding wave between the two wafers while said first and second pressures are maintained.

2. Bonding process according to Claim 1, in which the grooves (4) are arranged in the form of an orthogonal grid over the entire surface (6) of the chuck (2).

3. Bonding process according to Claim 1, in which the grooves (4) are arranged in the form of concentric annular grooves centred on the centre of the chuck (2).

4. Bonding process according to any one of Claims 1 to 3, in which the grooves (4) are uniformly distributed over the entire surface (6) of the chuck (2).

5. Bonding process according to any one of Claims 1 to 3, in which the grooves are more closely spaced in one zone of the surface of the chuck than the grooves in the rest of the surface of the chuck.

6. Bonding process according to any one of Claims 1 to 5, in which the pressure difference between said first and second pressures (PI, P2) is greater than or equal to 3 mbar. ^s

7. Bonding process according to Claim 6, in which said pressure difference is between 3 and 10 mbar.

8. Bonding process according to any one of Claims 1 to 7, in which the chuck (2) heats the first wafer (10) at least during the, contacting step and the step of initiating the propagation of the bonding wave.

9. Bonding process according to any one of Claims 1 to 8, which furthermore comprises:

- annealing (S25) the multilayer structure (20) resulting from the direct bonding of the first and second wafers; and

- thinning (S30) the first wafer (10) or the second wafer (16).