Etching of Silicon Nitride in 3D NAND Structures

3D NAND structures present a unique difficulty in semiconductor device manufacturing.  One method of production consists of cylindrical structures made of alternating layers of silicon dioxide (SiO₂) and silicon nitride (SiN) with a core of SiO₂.  As part of the device manufacturing process the SiN is etched away, leaving the SiO₂ core with disc-shaped SiO₂ fins.  Due to material selectivity requirements of etching only the SiN while leaving the SiO₂ unchanged, the process of choice is often immersion in a high-concentration hot phosphoric acid (H₃PO₄) bath with a silica additive to inhibit SiO₂ etching, which with proper tuning can have selectivity of greater than 100:1 for etching SiN:SiO₂.

Like all wet etch processes, uniformity of the process is a constant concern.  One of our goals in this paper is to show how we use experimental measurements and simulations to better understand what parameters have the most effect on the process, and how we can use that knowledge to improve the process.

Experimentally we have measured the etch rate of blanket SiN and SiO₂ wafers.  We fit the etch rate data to Arrhenius models to quantify the etch rate of SiN and SiO₂ films.  Simulations of the etch process on both blanket wafers and patterned wafers showed that the etch rate is the same and doesn’t depend on feature geometry.

In this paper we propose a simple chemical reaction and diffusion mechanism that explains the experimental observations of the etched structures.  The uniformity of the SiN etching between the top and the bottom of the 3D NAND stacks indicates that the reaction rate is much slower than the rate of diffusion of reactants and products into and out of the structure, so the etching is a reaction-controlled mechanism.  This indicates within-wafer uniformity and wafer-to-wafer uniformity variations as being primarily due to differences of temperature within the bath.

We also conducted simulations of the etching of the SiN in the 3D NAND structure and the liquid flow in the bath to model and explain this mechanism.  In these simulations we varied the rate of the surface reaction of H₃PO₄ etching SiN, and showed that for a fast reaction rate that the structure would etch from the top to the bottom as the overall reaction rate become diffusion limited, while for a slow reaction rate the entire structure would etch uniformly, showing it to be a reaction limited process.

Another issue with SiN etching in an H₃PO₄ bath is that the etch rate of SiO₂ depends highly on the concentration of silica, in fact the bath is generally pre-seeded with silica in order to achieve a lower SiO₂ etch rate, though if above a critical concentration deposition is known to occur.  Based on a study of silica chemistry, it may be possible for silica deposition to occur even when below the critical concentration.  This would be a film that would grow on a SiO₂ surface but would later dissolve back into solution.

We propose a mechanism that explains the formation and later disappearance of the growth film on the SiO₂ surface as a result of the etching of the SiN and the formation of silica byproducts into a sol/gel network.  Sol/gels have the unique property of their ability to form in an area of locally high flux, and then later dissolve over a much longer time scale.  The difference in forward and backward reaction rates of the sol/gel film formation and disillusion, respectively, are such that the system is never in equilibrium.  Thus, even though all the silica can dissolve in the phosphoric acid at equilibrium, the silica formed from the SiN etching forms into a sol/gel polymer network that only slowly dissolves after SiN etching has stopped.  We discuss the implications of this sol/gel oxide film for further process improvement.

With better theoretical understanding of the SiN etching and possible silica sol/gel formation and dissolution process, we now have tools to better optimize the phosphoric acid bath, improving both etching uniformity and necessary process time.