Fermi acceleration in a Fermi-Ulam model, consisting of an ensemble of particles bouncing between two, infinitely heavy, stochastically oscillating hard walls, is investigated. It is shown that the widely used approximation, neglecting the displacement of the walls (static wall approximation), leads to a systematic underestimation of particle acceleration. An improved approximative map is introduced, which takes into account the effect of the wall displacement, and in addition allows the analytical estimation of the long term behavior of the particle mean velocity as well as the corresponding probability distribution, in complete agreement with the numerical results of the exact dynamics. This effect accounting for the increased particle acceleration -Fermi hyperacceleration-is also present in higher dimensional systems, such as the driven Lorentz gas. In 1949 Fermi [1] proposed an acceleration mechanism of cosmic ray particles interacting with a time dependent magnetic field (for a review see [2]). Ever since, this has been a subject of intense study in a broad range of systems in various areas of physics, including astrophysics [3,4,5], plasma physics [6,7], atom optics [8,9] and has even been used for the interpretation of experimental results in atomic physics [10]. Furthermore, when the mechanism is linked to higher dimensional timedependent billiards, such as a time-dependent variant of the classic Lorentz Gas, it has profound implications on statistical and solid state physics [11]. Several modifications of the original model have been suggested, one of which is the well-known Fermi-Ulam model (FUM) [12,13,14] which describes the bouncing of a ball between an oscillating and a fixed wall. FUM and its variants have been the subject of extensive theoretical (see Ref.[13] and references therein) and experimental [15,16,17] studies as they are simple to conceive but hard to understand in that their behavior is quite complicated. A standard simplification [13] widely used in the literature, the static wall approximation (SWA), ignores the displacement of the moving wall but retains the time dependence in the momentum exchange between particle and wall at the instant of collision as if the wall were oscillating. The SWA speeds up time-consuming numerical simulations and allows semi-analytical treatments as well as a deeper understanding of the system [13,18,19,20,21]. However, as shown by Einstein in his treatment of the Brownian random walk [22], taking account of the full phase space trajectory (instead of the momentum component only) is essential for the correct description of diffusion processes. More recently, in the context of diffusion in the deterministic FUM, Lieberman et al have shown that one has to employ both canonical conjugate variables (position and momentum) in order to obtain the correct momentum distribution in the asymptotic steady state [20]. The present work shows that even in the absence of an asymptotic steady state the diffusion in velocity space is deeply affected by the location of th...