A case study is presented based on the use of high throughput experimentation (HTE) for the discovery of new memory materials.
BY LARRY CHEN, MARK CLARK, CHARLENE CHEN, SUSAN CHENG and MILIND WELING, IMI Inc., San Jose, CA
The ever increasing demands for data translate into more sophisticated and specific thin film requirements for semiconductor materials. Each film layer has to not only demonstrate desired film properties, but also show good interfacial behavior with neighboring layers to contribute to the performance of the whole film stack or device. As a result, modern thin film material systems are including more elements from the periodic table with more complex compositions. The demand for short time to market has also increased, making the development of new materials even more difficult. In this paper, we present a case study of using high throughput experimentation (HTE) for the discovery of new memory materials. By using a combinatorial approach of sputtering technology, HTE can be applied to PVD chalcogenides and other materials targeted at memory semiconductors.
Ever since the deposition of materials by magnetron sputtering was introduced by F. M. Penning, the technology has become a major method for industrial thin film deposition, which typically generates dense, hard, and robust thin film materials at relatively low production cost. The technology has been applied to major industries such as semiconductors, photovoltaics, optical coatings, displays, hard mechanical coatings, and so on. However, optimizing the magnetron sputtering processes has always been challenging to process and hardware design engineers, since material properties like density, crystalline structure, grain size, optical indices of a deposited film strongly depend on various process parameters, such as power, pressure, substrate temperature, sputter gas type, plasma type, sputter source to substrate distance, substrate bias, and pumping throughput. Additionally, the material properties heavily depend on the underlying layers, including the chosen substrate, below a film stack due to a texture effect in film structure and a formation of interfacial layers which comes from the intermixing of both materials. All the above parameters contribute to increasing the level of complexity of the development.
The semiconductor industry is constantly searching for new materials with unprecedented physical, optical, electrical, and mechanical properties, not only as a single film but also as a component of complex featured film stacks or functioning devices. This requires exploration of new materials not limited to pure or binary systems, but to ternary, quaternary systems and beyond. A very efficient solution to cope with the increasing complexity of development and the demand for short development time is a combinatorial approach.
The combinatorial approach can be defined as a process that couples the capability for parallel production of large arrays of diverse materials together with different high-throughput measurement techniques for various intrinsic and performance properties supported by data analytics for identifying lead materials . For magnetron sputtering technology, the optimization of process param- eters has to be included as a major component of combinatorial approach. Considering all the multi-dimensional space of the development mentioned above, the combinatorial approach can be an excellent and efficient way of developing new materials in magnetron sputtering in terms of cost and time.
HTE methodology for PVD materials discovery
Platform Considerations As all process parameters in magnetron sputtering are somewhat correlated, it has been challenging for process engineers to come up with fully optimized process parameters for thin film production. In addition, semiconductor production facilities are typically optimized for consistent, efficient, high volume production of a single product at a time, and not for a wide range of simultaneous experiments. These factors make it challenging for memory manufacturers to test multiple materials, conditions and devices in an efficient manner, and without compromising either data quality or production throughput.
IMI’s high throughput experimentation (HTE) platform is set up for accelerated experimentation. Its combina- torial PVD tool typically has four sputter guns and one additional port at the center. All sputter guns can be equipped with various types of target materials including chalcogenides, puremetals, oxides, and nitrides, and each sputter source can be operated by different plasma modes independently, such as direct current (DC), pulsed direct current (PDC), and radio frequency (RF) with the ability to co-sputter with all four guns. The additional port at the center can be equipped with an ion beam source for ion beam assisted deposition, or ion beam cleaning, or an additional sputter gun which enables five gun co-sputtering operation. Process parameter windows can cover larger regimes than most production tool process parameters (Table 1).
FIGURE 1 shows an example of a multi-target sputter chamber capable of controllably forming a variety of compounds in an array across a 300 mm substrate and an example substrate shown at right. The materials can also be deposited on a die-to-die basis (not shown) over a 300mm wafer test vehicle for direct device testing without the need for patterning. The effectiveness of the combinatorial screening can be increased by guiding the selection of material compositions using both semi-phenomenological and DFT-based modeling, as well as relating the experimental data to the results obtained from simulated annealing using ab-initio molecular dynamics and further DFT analysis of the simulated quasi-amorphous structures.
Two different methods can be used to deposit the combinatorial films of interest: site isolated spot and gradient approaches. For the site isolated spot approach, multiple numbers of spots were deposited on a substrate. Each individual spot represents a split condition from a design of experiment (DOE). Film composition can be controlled through the co-sputter of guns, which are equipped with targets consisting of different materials. Also, the process condition of each spot can be varied through the process parameter settings. All deposition conditions and procedures are fully automated.
In the gradient approach, non-uniform film in terms of composition and thickness is intentionally generated on top of a substrate by co-sputtering through an open large area aperture. A semi-empirical model is used for the control of non-uniformity. The modeling also helps in controlling the film composition throughout a target’s lifetime. In this approach, composition gradients and the thickness gradients can be generated by a single film deposition on a substrate. Theoretically, an infinite number of variations can be analyzed within a film, which is only limited by the spatial resolution of metrologies.
Characterization and device performance
Once films have been deposited via PVD, characterization can be carried out, including testing of physical, optical and electrical parameters. These can range from general film characteristics including composition, thickness and crystallinity, to device-specific electrical parameters such as leakage, threshold voltage, and On/ Off ratio.
Measuring and analyzing large numbers of data generated from HTE methodology can be time- consuming. By using the automated metrology tools and a unified database system, measurements and analysis steps can be expedited to limit bottlenecks and deliver data most efficiently. A multi-stage approach can also help to prioritize and focus experimental resources on the most promising candidates.
HTE vs traditional methods
Key benefits of the HTE approach include the expedited learning cycle, cost reduction, and improved data quality. For semiconductor applications, a single 200mm or 300mm wafer can hold more than 30 splits, which can lead to a reduction in cycle of learning time (one device wafer instead of more than 30). Additionally, as all spots on a single wafer go through the same follow-up device fabrication steps together, data can be free from unexpected fluctuations of subsequent steps. Overall, the HTE approach can expedite the learning cycle by 5 ~ 10 times compared to single substrate based approach. A comparison of both HTE with traditional methods is summarized in Table 2.
A case study in NVM
New materials for memory elements such as non-volatile memory (NVM) selectors must meet a wide range of performance parameters (FIGURE 3 shows a typical memory cell with the selector element called out), in order to reduce sneak currents and manage variability in memory arrays.
Table 3 lists some of the key parameters desired in a memory selector material.
Of course, optimizing all of these parameters simultaneously in a single element or compound (and one that is practical for high volume memory manufacturing) is challenging. IMI’s HTE methodology enables rapid and simultaneous optimization of key trade-offs between performance, reliability and integration, in the quest for an ideal selector.
HTE for NVM selector materials
Use of a HTE methodology allows rapid screening of NVM selector candidate material compounds, compo- sitions and stacks. IMI has conducted multiple customer engagements in memory selector materials screening, and a typical experimental workflow is outlined in FIGURE 4, showing progression from PVD deposition, through physical and electrical characterizations of films and devices.
This experimental process can be carried out multiple times, through subsequently more advanced stages on a fewer number of samples, as promising candidates are narrowed down and further optimized. FIGURE 5 shows a possible strategy for testing a series of candi- dates through three different stages. In the earlier stages, a wide range of options could be screened quickly, but the more extensive (and time consuming) characterization and analysis can be saved for later stages, when only the best performing candidates are already selected. This enables the best use of deposition and testing resources, leading to optimal results in an efficient timeframe.
Fast and high-quality experimental results
IMI has extensive experience in working both on dynamic random access memory (DRAM) as well as NVM materials. In DRAM, the company has worked on development of dielectric, electrode and interface layer materials. IMI’s process engineers, materials scientists and electrical engineers work upfront with a customer on the design of experiments to ensure the delivery of rapid cycles of learning with the most efficient use of resources.
A typical customer project might range between a few months up to a year or more, encompassing hundreds or even thousands of different experiments. In NVM selectors alone, IMI has conducted:
• 2500+experiments on Metal Chalcogenides
• 2000+ experiments on MIEC
• 1000+experiments on Transition Metal Oxides
High throughput experimentation can offer rapid, high quality materials data when effectively applied to PVD memory selector development. However it does require an advanced platform, and a facility and team experienced in efficient deposition and testing of the materials and devices. Materials and device expertise is also helpful in managing and optimizing the experimental workflow for maximum efficiency and high quality data.