by Debra Vogler, senior technical editor
Brion Technologies is unveiling its second-generation system, Tachyon 2.0, at this week’s SPIE Advanced Lithography Conference (Feb. 26, San Jose, CA). Still using Xilinx FPGAs (current generation) to achieve hardware acceleration, the company claims that a single Tachyon 2.0 system rack can replace the production capacity of four of the first generation racks.
“In addition, the modeling power, measured by the number of calculations we do in the modeling of optical lithography, goes up by a factor of four,” Jim Wiley, senior technical director at Brion, explained to WaferNEWS.
The need for computational speed as well as greater accuracy is becoming more significant as the industry deals with the challenges of low k1 imaging as it goes to 45nm and 32nm. The techniques used by lithographers to maximize resolution under low k1 conditions include extreme off-axis illumination, extreme annular illumination, and even dipole illuminations, which are being pushed out to the edge of the illuminator lens — all of these require more computing power.
Brion claims the new system (which is based on IBM Linux servers) provides a 4x improvement in modeling power as well as a 4x increase in speed for performing OPC and OPC verification. To get increased process window coverage, computational lithography tools need more computing power to handle more convolution kernels and a larger optical interaction range.
According to Wiley, the company gave strong consideration to other kinds of accelerator chips, i.e., gaming chips, video chips, graphics chips, and general purpose accelerators. “We felt that going all the way to an FPGA, which is the most extreme form of acceleration, was still the best choice,” Wiley said.
At the 45nm half-pitch, the Tachyon 2.0 system has an optical interaction range of 4-microns, compared to a 1-micron range for the Tachyon 1.0 at the same node, according to the company’s specifications. For 45nm designs, the new system surveys an area up to 8-microns in diameter. It also improves modeling power by supporting more optical simulation kernels. “We go from a typical 64 complex kernels for calculations with the Tachyon 1.0, to 256 kernels in the Tachyon 2.0,” said Wiley. — D.V.