Technical Insight
ICSCRM underlines good progress in SiC technology
This year's ICSCRM highlighted the significant steps that have been taken recently towards broadening the range of SiC devices available off the shelf. Jon Newey reports.
The recent International Conference on SiC and Related Materials (ICSCRM 2003) illustrated very well that two years is a long time for a fledgling technology. Many of the 500 delegates gathered in Lyon, France, over October 5-10, 2003 were at the last ICSCRM in Japan in 2001, when Infineon and Cree had just begun to compete as the first two companies to commercialize SiC Schottky diodes. Since then, successive generations of SiC Schottkys have been released, handling greater levels of power. Rohm is just about to enter into SiC Schottky production, and Cree and Rockwell Scientific have released commercial SiC MESFET products. Behind the scenes there has been rapid progress in SiC R&D labs. Bulk and epitaxial material quality has improved, something that device developers have taken full advantage of, and new processing strategies are helping to overcome old barriers to progress.
Given the still limited commercialization of SiC devices, the fact that so many delegates attended illustrated the strong level of funding being invested in SiC R&D. The funding given to SiC and other wide-bandgap semiconductors by the US Department of Defense was perhaps reflected in the fact that more than 100 delegates were from the US, with Japan having a similar number, and France and Germany accounting for 90 and 60 respectively. As well as established SiC players such as Cree, II-VI and Okmetic, the exhibit featured some new faces, including SemiSouth, Bandgap Technologies and Dow Corning, which acquired SiC substrate manufacturer Sterling Semiconductor in January 2003.
Micropipes are historyCree s Don Hobgood gave the opening talk of the conference, bringing the audience up to date on the state of the art in SiC substrates, the impact this has had on devices, and the issues that are now being tackled. Micropipes have been the main issue for substrates for some time, but recent progress is changing this. Cree s best result was a 75 mm 4H wafer with just 10 micropipes. To put this in perspective, 10 A Schottky diodes fabricated using 75 mm wafers with a micropipe density of 0.45/cm2 would have a yield of 97%, and the yield for larger 100 A devices would be 79%. The current best micropipe density for 100 mm 4H-SiC wafers is 22/cm2. While 10 A device yield on such a wafer would drop to 67%, it would still provide 970 good devices, far more than the very best 3 inch material. The economics of manufacturing may well dictate that large-area wafers with more micropipes are good enough for many device types, although material scientists might not be happy until micropipes are a thing of the past.
Given the problems of forward voltage drift in bipolar devices (see "Bipolar devices", next page), basal plane dislocations are replacing micropipes as the enemy on the battlefield. The density of these defects in current best material is 103-104/cm2, and their presence in epilayers has been linked to the formation of stacking faults in the thick epitaxial drift layers of bipolar devices under stress conditions.
The best-performing AlGaN/GaN HEMTs have been those fabricated on SiC substrates thanks to their high thermal conductivity and closer lattice match to GaN than sapphire or Si substrates. Highlighting the link between good substrates, high-quality epitaxy and device performance, Hobgood described some impressive figures from an AlGaN/AlN/GaN HEMT grown on a 75 mm semi-insulating SiC wafer. The sheet resistance of the epi structure grown onto a Fe-doped buffer was 280 Ω/ with less than 1% standard deviation across the wafer. The HEMT, fabricated using a field plate on the gate electrode and an AlN barrier layer above the GaN channel, operated at 4 GHz and had a current density of 20.4 W/mm with 51% PAE at 82 V (figure 1). If GaN-based devices can handle such power density, why is there such interest in SiC? John Palmour, Cree s executive VP of advanced devices, put the case for SiC succinctly in the Industrial News session: "SiC is very reliable at high powers, but the reliability curves for GaN devices are still horrible. We are interested in both materials, but in the meantime we need to sell something."
MOSFETs and channel mobilityThe development of SiC MOSFETs has been hampered by poor oxide reliability and low channel mobility. The low channel mobility is mainly the result of a high density of interface states near to the conduction band, which trap or scatter the carriers. The subject of MOSFETs and the oxide/SiC interface was widely discussed and fortunately reflected some good progress since ICSCRM 2001.
Dietrich Stephani of SiCED in Germany gave a plenary talk reviewing the progress, outstanding issues and current state of MOSFET development at SiCED, which has resulted in the demonstration of a vertical device with a blocking voltage of 3 kV. In blocking mode, the normally off device showed a leakage current of
Given the still limited commercialization of SiC devices, the fact that so many delegates attended illustrated the strong level of funding being invested in SiC R&D. The funding given to SiC and other wide-bandgap semiconductors by the US Department of Defense was perhaps reflected in the fact that more than 100 delegates were from the US, with Japan having a similar number, and France and Germany accounting for 90 and 60 respectively. As well as established SiC players such as Cree, II-VI and Okmetic, the exhibit featured some new faces, including SemiSouth, Bandgap Technologies and Dow Corning, which acquired SiC substrate manufacturer Sterling Semiconductor in January 2003.
Micropipes are historyCree s Don Hobgood gave the opening talk of the conference, bringing the audience up to date on the state of the art in SiC substrates, the impact this has had on devices, and the issues that are now being tackled. Micropipes have been the main issue for substrates for some time, but recent progress is changing this. Cree s best result was a 75 mm 4H wafer with just 10 micropipes. To put this in perspective, 10 A Schottky diodes fabricated using 75 mm wafers with a micropipe density of 0.45/cm2 would have a yield of 97%, and the yield for larger 100 A devices would be 79%. The current best micropipe density for 100 mm 4H-SiC wafers is 22/cm2. While 10 A device yield on such a wafer would drop to 67%, it would still provide 970 good devices, far more than the very best 3 inch material. The economics of manufacturing may well dictate that large-area wafers with more micropipes are good enough for many device types, although material scientists might not be happy until micropipes are a thing of the past.
Given the problems of forward voltage drift in bipolar devices (see "Bipolar devices", next page), basal plane dislocations are replacing micropipes as the enemy on the battlefield. The density of these defects in current best material is 103-104/cm2, and their presence in epilayers has been linked to the formation of stacking faults in the thick epitaxial drift layers of bipolar devices under stress conditions.
The best-performing AlGaN/GaN HEMTs have been those fabricated on SiC substrates thanks to their high thermal conductivity and closer lattice match to GaN than sapphire or Si substrates. Highlighting the link between good substrates, high-quality epitaxy and device performance, Hobgood described some impressive figures from an AlGaN/AlN/GaN HEMT grown on a 75 mm semi-insulating SiC wafer. The sheet resistance of the epi structure grown onto a Fe-doped buffer was 280 Ω/ with less than 1% standard deviation across the wafer. The HEMT, fabricated using a field plate on the gate electrode and an AlN barrier layer above the GaN channel, operated at 4 GHz and had a current density of 20.4 W/mm with 51% PAE at 82 V (figure 1). If GaN-based devices can handle such power density, why is there such interest in SiC? John Palmour, Cree s executive VP of advanced devices, put the case for SiC succinctly in the Industrial News session: "SiC is very reliable at high powers, but the reliability curves for GaN devices are still horrible. We are interested in both materials, but in the meantime we need to sell something."
MOSFETs and channel mobilityThe development of SiC MOSFETs has been hampered by poor oxide reliability and low channel mobility. The low channel mobility is mainly the result of a high density of interface states near to the conduction band, which trap or scatter the carriers. The subject of MOSFETs and the oxide/SiC interface was widely discussed and fortunately reflected some good progress since ICSCRM 2001.
Dietrich Stephani of SiCED in Germany gave a plenary talk reviewing the progress, outstanding issues and current state of MOSFET development at SiCED, which has resulted in the demonstration of a vertical device with a blocking voltage of 3 kV. In blocking mode, the normally off device showed a leakage current of
