堆叠结构的晶圆及其减薄方法

Abstract— Wafer-to-wafer 3D integration has a potential to  minimize the Si thickness, which enables us to connect  multiple wafers with significantly scaled through-Si vias. In  order to achieve this type of 3D structure, backside thinning  is a key step. Conventional mechanical grinding is known as  the best way to remove bulk Si in terms of cost of ownership  (CoO). However, mechanical damage such as induced  dislocations needs to be removed after extreme thinning to  avoid a serious impact on the device performance. CMP  shows the best performance in terms of roughness with a  significantly flat surface with only atomic step roughness.  Furthermore, the existing mono-vacancies are as low as for a  bulk Si substrate. However the total thickness variation  (TTV) worsens as more Si is removed. The dry etch process  enables a faster etch rate than CMP and wet etching.  Furthermore, the mono-vacancy/damage layer after dry  etching is equivalent to that achieved when combined with  CMP. The combination of CMP and dry etch enables us to  achieve extreme thinning of active device wafers (<5 μm)  with minimal roughness, no damage layer (mono-vacancy)  and no edge delamination.  Keywords-component; Edge-trim, Grinding, CMP, Dry  etch, wafer-to-wafer bonding,

 INTRODUCTION 

Wafer-to-wafer stacking potentially offers a solution  for thin wafer handling issues allowing the removal of  most of the Si from the top wafer. The extremely thin Si (<  5 μm) can extend the scaling of via-last TSVs, due to the  minimized pitch and dimension (pitch < 2 μm, dimension  1 x 5 μm). The via-last TSV formation is significantly  influenced by the thinning performance, which brings new  challenges in the thinning process [1]. A smooth surface  and less damage into the Si are mandatory to achieve a  successful backside TSV process. Accurate thickness  control and minimal total thickness variation (TTV) across  the whole wafer are required for the via etch process [2].  In particular, TTV is much more critical for the range of  thin Si. In addition, any mechanical failures occurring at  the very wafer edge, an area of the wafer typically  excluded from inspection, may also impact multiple waferto-wafer bonding. Furthermore, vacancy-type of atomic  defects need to be analyzed using depth profilinge for the  extreme thinning, since the remaining Si is limited. Thus, adeeper understanding of micro and macro surface  characterization is essential for extreme thinning.   In this paper the characterization of different extreme  Si wafer thinning processes are discussed on permanently  bonded wafers. Several thinning methods, such as CMP  and plasma dry etch are compared as subsequent processes  of grinding. The wafer scale topography such as TTV,  mean thickness control and wafer bowing will be  discussed. The wafer inspections were performed up to the  wafer bevel. In addition, nano-scale characteristics, such  as roughness, mono-vacancies and damage in/to Si will be  investigated by using alternative ways of surface detection.

EXPERIMENTAL

  The thinning processes were performed on the top  wafer of permanently bonded wafer stacks. The permanent  wafer bonding was performed on 300 mm wafers. Oxide  CMP is used to planarize and to smooth the oxide layer.  Prior to bonding, the pairing wafers are treated with a  combination of plasma activation and a DI water clean.  Wafer bonding takes place at room temperature with  atmospheric pressure in a clean room ambient. Then an  annealing cure was used to improve the adhesion strength.  The edge-trim process was performed using a dicing saw  tool DFD6860 from Disco. The grinding step was carried  out on a Disco DGF8560 series in-feed grinder which  consists of rough and fine grinding wheels. The wafer  nano-tomography was measured by a wafer automated  inspection system. The wafer edge inspections were done  in a KLA-tencor CIRCL-AP.  The dry etching for Si thinning was performed by an  SPTS Rapier XE system with Near Infra-Red (NIR)  interferometer in-situ endpoint detection.

                                                                                       Figure 1. Wafer edge shadow images taken after edge-trim and  grinding for different edge-trim approaches

A comparison of the damage into the Si was made  between edge-trim before and after bonding. In order to  compare the edge-trim impact cross-sectional TEM images  were taken at the wafer edge after grinding. As the edgetrim depth after bonding is much deeper than for edge-trim  before bonding, a blade with bigger diamonds is suitable  for edge-trim after bonding. Figure 2 shows TEM images  of dielectric bonded wafers after grinding using edge-trim  before bonding (a)-(b) and after bonding (d)-(e).  Dislocations and stacking faults in the {111} plane can be  seen in the damaged regions in both samples. The presence  of acontinuous amorphous Si layer was also observed at  the wafer edge. The Si sidewall using edge-trim after  bonding has a higher roughness, deeper defects (~500 nm)  and thicker and continuous amorphous Si layer. This  might be due to the transfer of a significantly big  indentation and heat energy caused by the diamond  abrasive process. The amorphous Si and different phases  of Si are also detectable by using micro-Raman  spectroscopy. Figure 2 (c) and (f) shows Raman spectra  taken from the sidewall of the edge-trimmed Si. A lot of  peaks can be distinguished for the case of an edge-trimmed  sidewall. This indicates the temporary occurrence of  significant high local stresses during blade dicing [3,4].  This does not create an immediate integration failure but  these defects might represent the initiation of a fragile  wafer edge.