The old question of "How long does it take to tunnel through a barrier?" has acquired new urgency with the advent of techniques for the fabrication of semiconductor structures in the nanometer range. For the restricted problem of tunneling in a scattering configuration, a coherent picture is now emerging. The dwell time ${\tau }_D$ has the status of an exact statement of the time spent in a region of space, averaged over all incoming particles. The phase times ${\tau }_T^{\varphi }$ and ${\tau }_R^{\varphi }$ are defined separately for transmitted and reflected particles. They are asymptotic statements on completed scattering events and include self-interference delays as well as the time spent in the barrier. Consequently, neither the dwell time nor the phase times can answer the question of how much time a transmitted (alternatively, reflected) particle spent in the barrier region. Our discussion of this question relies on a few simple criteria: (1) The average duration of a physical process must be real. (2) Since transmission and reflection are mutually exclusive events, the times ${\tau }_T$ and ${\tau }_R$ spent in the barrier region are, if they exist, conditional averages. Consequently, they must obey the identity ${\tau }_D=T{\tau }_T+R{\tau }_R$, where $T$ and $R$ are the transmission and reflection probabilities, respectively. The existence of this identity distinguishes tunneling in a scattering configuration from tunneling out of a metastable state. (3) Any proposed ${\tau }_T$ and ${\tau }_R$ must meet every requirement that can be constructed from ${\tau }_D$. On the basis of (2), the naively extrapolated phase times, as well as the Büttiker-Landauer time, must be rejected. The local Larmor times, as introduced by Baz', satisfy (2), but not every criterion of type (3). The local Larmor clock is therefore unreliable. The asymptotic Larmor clock shows the phase times, as it should. Finally, the inverse characteristic frequency of an oscillating barrier cannot always be defined. It is shown not to represent the duration of the tunneling process. This leaves the dwell time and the phase times as the only well-established times in this context. It also leaves open the question of the length of time a transmitted particle spends in the barrier region. It is not clear that a generally valid answer to this question exists.

DOI:https://doi.org/10.1103/RevModPhys.61.917