An Alternative Approach to Measure Alpha Particle Induced Soft Error Rate for Flip-Chip Packaged SRAM Devices

Abstract

Alpha particles emission from radioactive impurities, found in the chip and packaging materials in low concentration, poses a serious threat to semiconductor devices’ reliability. To evaluate a device sensitivity against alpha particles, accelerated Single Event Effect (SEE) tests are conducted using alpha source, which emits particles just over 5 MeV. For devices mounted in recommended packaging, for alpha particle SEE testing, the source is placed directly above the die to expose the active circuit to the alpha radiation. However, relentless downscaling and package miniaturization trends have driven the demand of developing increasingly complex packaging – i.e. smaller, thinner with tremendous input/output pins count per chip. Flip-chip Ball Grid Array (BGA) packaged devices meet all these challenges rather efficiently, but their testing against alpha particle is a big challenge. The range of particles emitted from alpha source is very short (less than 30 µm in silicon), which prevents their penetration till the active circuit of such devices — from either side of the chip. Accordingly, to test such devices, the part of package and chip is chemically etched-away from backside to allow particles to arrive at the active circuit to induce upsets. In this dissertation, high energy alpha particles irradiation has been proposed as a potential alternative approach to assess Single Event Upset (SEU) sensitivity for flip-chip packaged SRAM devices. The method uses long-range, high energy alpha particles, directed from the backside of silicon die, which can easily penetrate the entire substrate and deposit charge in the sensitive region of the device to induce upsets. The LET (Linear Energy Transfer) at active circuit is varied by changing the incident energy at die surface from the backside. Alpha particle incident energy, which mimics low energy (≈ 5 MeV) at the active circuit from the backside of die for given device substrate thickness, is determined using SRIM — which is an Monte Carlo program. The cross-section is measured at various alpha particle high energy, including the energy corresponding to Bragg Peak and energy which mimics ≈ 5 MeV at the active circuit. The SEE experiments are performed on 14nm FinFET and 55 nm CMOS SRAM devices, assembled in flip-chip and wire-bonded structures, respectively, using proposed backside and traditional top-side testing, respectively. CREME-MC, a Geant4-based Monte-Carlo transport code, is used to show the energy deposition in sensitive volume and resulting SEU cross-section induced by high energy alpha particle irradiation. The correlation between the traditional and proposed methods’ results suggests that the proposed method may be used to assess flip-chip bonded SRAM devices sensitivity, without having to etch away the part of the package and chip.