What makes a good precursor in atomic layer deposition isn’t clear.
Equipment and materials suppliers often talk about the fragmentation of integrated circuit processing. While the number of manufacturers has gone down, the diversity of the underlying semiconductor market has increased.
Low-power processors for mobile devices, non-volatile memory for solid state disks, and dedicated graphics processors all have different requirements from the traditional industry drivers of high-performance, general-purpose microprocessors and DRAMs. Often, this means that instead of having a single process of record for, say, hafnium oxide deposition, there are many, depending on the needs of the specific device. Even a seemingly simple question — “What is the best precursor for hafnium ALD?” — turns out to have a surprisingly subjective answer, depending on the particular manufacturer’s requirements as much as on a particular compound’s objective characteristics.
A good precursor is volatile enough to supply sufficient hafnium for reasonable process throughput, but stable enough to avoid undesirable reactions in the process tool’s precursor transport system, on chamber surfaces, or in the chamber effluent. It must be reactive with its co-precursors — typically water or ozone for hafnium oxide deposition — and the resulting film should be free from particles and contaminants. For dielectric applications, leakage and dielectric constant are the critical material parameters.
Yet specific processes inevitably impose tradeoffs and force manufacturers to prioritize specific characteristics. For example, the gate dielectric in a logic process is a very thin layer, only a few nanometers thick. Leakage and contamination requirements are critical to overall circuit performance. On the other hand, logic processes generally use single-wafer or small batch deposition chambers. The total area that must be covered in a single pass is relatively small, and so the volume of precursor required is relatively low.
In a DRAM process, in contrast, the deposition must fill a deep trench, with an aspect ratio ranging from 40:1 to as high as 70:1. Typically, deposition takes place in a large batch reactor. The area to be covered is much larger, as is the sheer volume of precursor required per ALD cycle. Meanwhile, film quality requirements are generally less stringent.
These differing priorities are reflected in the precursors chosen by device manufacturers. In the logic sector, according to Sven Van Elshocht, manager of Imec’s thin film deposition group, manufacturers have almost universally adopted HfCl, a solid material, oxidized with water. Being a solid, HfCl has relatively low volatility, but it yields high quality films with superior leakage characteristics and very low contamination. Though early ALD research identified chlorine contamination as a potential concern, it turns out that carbon contamination from metal-organic precursors poses more serious challenges for actual processes. HfCl is also stable up to more than 600º C, offering a large window for process optimization.
For memory applications, the picture is somewhat murkier. Here, the low volatility of HfCl is a significant disadvantage. Simply volatilizing enough precursor at once to coat the large surface area required by a DRAM process is challenging. Memory manufacturers have gravitated toward metal-organic liquid precursors, oxidized with ozone.
In the early days of hafnium oxide use, that meant TEMAH (tetrakis[ethylmethylamido]hafnium), which gives a high deposition rate per cycle and reduces substrate dependence relative to HfCl. However, its relatively poor thermal stability led to a search for alternatives. Mohith Verghese, ASM International’s director of product marketing, said the majority of the industry now uses cyclopentadienyl-based compounds. The variation among manufacturers is significant, though. Metal-organic chemistries allow manufacturers to tune the ligands surrounding the molecule’s hafnium core in order to balance thermal stability, deposition rate, and other requirements.
As memory density continues to increase, more stringent leakage requirements are driving increased interest in solid precursors for memory, as well. Moreover, memory technologies other than DRAMs have their own unique requirements. NAND blocking oxides have similar requirements to DRAMs and use similar precursors. RRAMs, on the other hand, are very sensitive to oxygen content and film quality contributes to RRAM device lifetime. Thus, RRAMs are likely to use more logic-like hafnium deposition processes.
Often, the dielectric layer is not pure hafnium oxide, either, but a laminate, with zirconium or aluminum oxide layers, too. Each precursor for each layer must have compatible chemistry and process requirements with all the other layers. As hafnium oxide deposition processes mature, the material is finding additional applications, ranging from silicon nanowire structures to optical elements. As these applications emerge from laboratories, they too will place new demands on precursor chemistries.