Strained silicon devices: design and validation of a tensile testbench for strain-dependent electrical characterisation

Details

Supervisors

Raskin, Jean-Pierre

Faculty

Ecole polytechnique de Louvain

Degree label

Master [120] : ingénieur civil électricien, à finalité spécialisée

Abstract

enSilicon absorbs light only at wavelengths shorter than about 1100 nm, its indirect band gap of 1.12 eV placing the short-wavelength infrared out of reach of an unstrained device. First-principles calculations predict that a uniaxial tensile strain of 1 to 2 % along [110] lowers that gap by 100 to 200 meV, enough to extend absorption into that range. Reaching such a strain on a working device, with a strain state that is known rather than assumed, is the step this work addresses, using silicon p-n junctions carried by a flexible polyimide membrane, designed and fabricated at Winfab in an earlier thesis. The objective is twofold: to build and characterise a testbench capable of imposing that strain, and to apply it to the junctions themselves. The first part compares the methods available for straining silicon against the four requirements set out in Chapter 1, and retains uniaxial traction of a polycarbonate lamella as the closest match, the strain state being the one requirement it does not fully establish. The resulting testbench, built at UCLouvain’s LEMSC laboratory, reaches a reproducible uniaxial strain of 2.10 %, seven times the maximum reached in the four-point bending configuration used for comparison, with the strain read from a bonded reference gauge at every loading step. The second part applies the testbench to the p-n junctions it was built for. Simulations of the reverse characteristic predict a breakdown voltage falling by roughly 29 % per percent of strain and a leakage current rising by a factor of seven at 1 % strain. An unreliable electrical connection to the die confined the exploitable campaign to 0 to 0.50 % strain, over which no response distinguishable from measurement scatter could be resolved. The shape of the measured reverse current, a continuous rise with no generation plateau, points instead to a field-driven conduction mechanism not captured by the simulation, a question left for further work.