No abstract

There has been considerable interest recently, in the formation of the source drain junctions of metal oxide semiconductor transistors using solid phase epitaxy (SPE) to activate the dopants rather than a traditional high temperature anneal. Previous studies have shown that this method results in high dopant activation as well as shallow junctions (due to the small thermal budget). In this we study the effect the temperature of SPE regrowth has on the boron activation. We find that boron activation has a monotonically increasing dependence on the temperature. Significantly, we show that by carrying out the SPE regrowth at temperatures above 1050°C, it is possible to obtain active concentrations well above the electrical solubility limits.

The challenge of achieving maximal dopant activation with minimal diffusion has re-awakened interest in millisecond-duration annealing processes, almost two decades after the initial research in this field. Millisecond annealing with pulsed flash-lamps or scanned energy beams can create very shallow and abrupt junctions with high concentrations of electrically active carriers, but solutions for volume manufacturing must also meet formidable process control requirements and economic metrics. The repeatability and uniformity of the temperature cycle is the key for viable manufacturing technology, and the lessons from the development of commercial rapid thermal processing (RTP) tools are especially relevant. Advances in the process capability require a fuller understanding of the trade-off between dopant activation, defect annealing. diffusion and deactivation phenomena. There is a strong need for a significant expansion of materials science research into the fundamental physical processes that occur at the short time scales and high temperatures provided by millisecond annealing.

Due to integration concerns, the use of meta-stable junction formation approaches like laser thermal annealing (LTA), solid phase epitaxial regrowth (SPER), and flash annealing has largely been avoided for the 90nm CMOS node. Instead fast-ramp spike annealing has been optimised along with co-implantation to satisfy the device requirements, often with the help from thin offset spacers. However for the 65nm and 45nm CMOS node it is widely accepted that this conventional approach will not provide the required pMOS junctions, even with changes in the transistor architecture.In this work, we will compare junction performance and integratablity of fast-ramp spike, flash, SPER and laser annealing down to 45nm CMOS. The junction depth, abruptness and resistance offered by each approach are balanced against device uniformity, deactivation and leakage. Results show that the main contenders for the 45nm CMOS are SPER and flash annealing – but both have to be rigorously optimised for regrowth rates, amorphous positioning and dopant and co-implant profiles. From the two, SPER offers the best junction abruptness (<1nm/dec) with leakage suitable for low power applications, while the flash anneal has the benefit of higher solid solubility (>4E20at/cm3) and less transistor modifications. As expected, Ge and F co-implanted spike annealed junctions do not reach the 45nm node requirements. For full-melt LTA, poly deformation on isolation can be reduced but geometry effects result in unacceptable junction non-uniformity.

No abstract