Batch Czochralski Processing: Some Cold, Hard Truths
The Good and the Challenge
Batch CZ processing, as described at a high, conceptual level in the first of this series of articles, has been a fundamental technology, enabling the semiconductor industry for many decades [1]. Until now, the nature of Batch CZ processing has made it extensible to each, next larger wafer size, through its relatively low cost and ease of implementation.
Regardless, there have been difficulties with the Batch CZ method that the industry has had to work around or accept. As already described in the previous article, manually loading the crucible, particularly one for 450mm wafers, will be time intensive, as will be the time required to melt the polysilicon charge. However, two additional problems exist. This article will focus on the impact of oxygen impurities, and non-uniformity of the doping within the silicon ingot, both of which are natural consequences of the Batch CZ process.
The Problem
Oxygen Impurities
Though the crucible is often referred to as “quartz,” it actually is fused, amorphous silicon dioxide (SiO2), or silica. With the length of time the crucible is exposed to high heat in melting the large batch of polysilicon, a considerable amount of oxygen is naturally released. This oxygen gets into the crystalline structure, creating defects. Fortunately, much of the oxygen escapes at the surface of the melted polysilicon, but a non-trivial amount remains.
While there can be some beneficial aspects to oxygen contamination, such as increased material strength and gettering capabilities, there are many good reasons to eliminate or reduce (control) the amount and variation of oxygen content in the final ingot. Reducing the amount of convection in the melt (and commensurate mixing of the oxygen coming from the crucible walls) helps to resolve the problem. This can be accomplished by growing the ingot under magnetic confinement [2]. This requires a large magnetic field, which not only intensifies the complexity of the system, but also the system power requirements and potential health risks. The Batch CZ system with a magnetic field applied is conceptually shown in the movie in Figure 1. A real-world system with a transverse field is shown in Figure 2 [3].
Doping
Without going into the physics behind the need for doping, or impurities, to be added to the silicon ingot for it to function as a semiconductor, suffice to say that without the added, right impurities (such as boron), silicon doesn’t conduct electricity. Think of it like adding a little salt to ultra-pure water. Without the salt contaminant there is no electrical conductance. With salt contaminant, a charge will flow through the water. The mechanism behind conductance in doped silicon is different than water, and the doped silicon only “sort of” (semi) conducts, so the analogy can’t be taken too far.
However, one similarity in the water analogy is that when a material crystalizes, whether it be water, metal, or silicon, the crystalline lattice will not actually accommodate, or make room for, the contaminants very well. Either the contaminant will be pushed out during the freezing, become entrapped in the form of a defect in the crystal, or, ideally, will be integrated into the crystalline structure, or lattice. Taking advantage of the ability for impurities to be “pushed” out, freezing water from the top down has been used to create ultra-pure, clear ice for sophisticated cocktails [4]. Similarly the solidification of silicon into a crystalline structure also pushes out impurities so only a fraction of the material used for doping remains in the lattice.
An important feature of this process for Batch CZ, is that the impurities not accommodated into the crystal lattice are pushed back into the silicon melt, which increases the impurities’ concentration in the melt. The percentage of the impurity that is accommodated depends upon impurity type and availability. Therefore, as this concentration naturally and continuously increases into the melt, it also increases in the crystal, in effect, feeding on itself.
The end result for the silicon ingot, created with Batch CZ, is that the concentration of dopant, and resistivity of the silicon wafers cut from the ingot, varies along its length. Depending upon the specific application (memory, processor, MEMS), this variation will have a significant impact on an integrated circuit’s performance. Ultimately, the Batch CZ process often results in only a portion of the ingot having the desired usable electrical properties.
2nd Generation: Continuous CZ or CCZ
Scientists have determined that diluting the melt, with additional polysilicon as the batch process proceeds, mitigates the problem of dopant concentration variation [5]. This method is referred to as Continuous Czochralski, or CCZ.
While CCZ helps solve the problem of dopant concentration variation, simply adding chunks of polysilicon to the melt, continuously, while making the ingot, creates two problems: the physical disturbance on the surface of the molten polysilicon, and the disruption of the temperature gradient. Both problems are quite serious. Any disturbance at the liquid/solid interface, where the ingot is being created, will cause defects in the crystal (ingot). Various methods have been used to mitigate this, including using polysilicon granules, rather than chunks, to produce smaller “waves,” and making use of a double-walled crucible, feeding the polysilicon into the space between the walls. This double-walled crucible serves as a barrier to physical disturbances created in the liquid polysilicon thereby providing some thermal protection between the higher temperature, outer annulus where silicon is melted, and the inner melt, where a lower temperature is required for solidification. This is shown conceptually in Figure 3.
Figure 3. Continuous CZ Concept for 450mm
3rd Generation CZ: Shallow-Crucible CCZ
The natural progression is to extend this refill technique and remove the need for the initial batch fill required for CZ. This is possible by creating a shallow crucible, and keeping the melt at a constant level through continuous feed of pelletized silicon and dopant. Though there are variations on this theme [6], one such version is shown conceptually in Figure 4. The shallow crucible also mitigates the need for a magnetic field to reduce oxygen contamination.
Figure 4. Continuous CZ with Shallow Crucible
Ultimately, though, there is a still a battle between the higher temperature required to melt the polysilicon in the outer annulus and the lower temperature required at the ingot interface, for crystallization. This push-pull combat creates a suboptimal situation where neither temperature is ideal, ultimately reducing the speed at which the ingot can be pulled from the melt while maintaining minimal defect density.
The Ultimate Wafer
The next article will describe a CCZ technique that decouples the melt vs. crystallize temperature conflict. Surprisingly, the technique has been used in production for decades, though in a different industry. However, it also has room for improvement. The original inventors of the method, in applying further enhancements to producing 450mm silicon wafers, may have found a path to creating the ultimate wafer.
References
[1] G. K. Teal, “Single Crystals of Germanium and Silicon—Basic to the Transistor and the Integrated Circuit,” IEEE Trans. Electron Dev. ED-23:621 (1976).
[2] P. Gungai, “Tailoring Oxygen Concentration Distribution in 300mm Czochralski Crystal of Pure Silicon using Cusp Magnetic Field,” Seventh International Conference on CFD in the Minerals and Process Industries CSIRO, Melbourne, Australia, 9-11 December 2009.
[3] R. N. Thomas, H. M. Hobgood, P. S. Ravishankar, and T. T. Braggins, “Melt Growth of Large Diameter Semiconductors: Part I,” Solid State Technol. 33:163 (April 1990).
[4] http://www.wintersmiths.com/collections/shop/products/the-ice-baller
[5] CCZ Patent: http://www.google.com/patents/US5492078
[6] Shallow Crucible CCZ Patent: http://www.google.com/patents/US5485802