Current filament (CF)-related double-hysteresis I-V behavior and holding current, I <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">HOLD</sub> ,are analyzed using experiments and 3-D technology computer-aided design (TCAD) simulation in silicon-controlled rectifiers (SCR) for system-level electrostatic discharge (ESD) protection. Our 3-D TCAD methodology uses up and down quasi-dc current sweeps to reveal a memory effect in the current density distribution along the device width. I <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">HOLD </sub> is related to the smallest possible CF where the self-sustaining SCR action takes place during down current sweep. I <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">HOLD</sub> exhibits a nontrivial dependence on device width, depending on whether a CF is created or not. Analyzing devices of different layouts shows that I <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">HOLD </sub> values determined from experiments and 3-D TCAD are almost layout-independent and substantially lower than those evaluated from 2-D TCAD. I <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">HOLD </sub> calculated by 3-D TCAD in edge-terminated devices is higher than that in 3-D structures obtained from simple width-extended 2-D doping profiles. The use of latter devices, thus, simplifies the 3-D TCAD I- V analysis and provides a safe margin for I <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">HOLD </sub> prediction. The work is relevant for designing the latch-up immunity of ESD protection devices, and it also shows that conventional 2-D TCAD can provide unwanted overestimation of I <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">HOLD</sub> .