The discovery rate of optical transients will explode to 10 million public alerts per night once the Vera C. Rubin Observatory’s Legacy Survey of Space and Time comes online, overwhelming the traditional physics-based inference pipelines. A continuous-time forecasting AI model is of interest because it can deliver millisecond-scale inference for thousands of objects per day, whereas legacy MCMC codes need hours per object. In this paper, we propose SELDON, a new continuous-time variational autoencoder for panels of sparse and irregularly time-sampled (gappy) astrophysical light curves that are nonstationary, heteroscedastic, and inherently dependent. SELDON combines a masked GRU-ODE encoder with a latent neural ODE propagator and an interpretable Gaussian-basis decoder. The encoder learns to summarize panels of imbalanced and correlated data even when only a handful of points are observed. The neural ODE then integrates this hidden state forward in continuous time, extrapolating to future unseen epochs. This extrapolated time series is further encoded by deep sets to a latent distribution that is decoded to a weighted sum of Gaussian basis functions, the parameters of which are physically meaningful. Such parameters (e.g., rise time, decay rate, peak flux) directly drive downstream prioritization of spectroscopic follow-up for astrophysical surveys. Beyond astronomy, the architecture of SELDON offers a generic recipe for interpretable and continuous-time sequence modeling in any time domain where data are multivariate, sparse, heteroscedastic, and irregularly spaced.

Recent studies of silent data errors (SDEs) in mega-scale datacenters indicate that SDEs are caused by small delay defects that escaped detection by manufacturing tests or occurred during the lifetime of the system. A small delay defect is detected by a test that propagates a transition through one of the longest paths that includes the defect site. Once the path is selected, for every gate or cell on the path, ATPG assigns off-path input values to enable the propagation of a transition through the path. For complex gates, such as AOI and XOR, there are multiple sets of possible assignments (or input stimuli) that the ATPG can use, with substantially different propagation delays. Existing test generation procedures do not consider these differences in propagation delays once a path is selected. We propose a new approach to ATPG for small delay defects, called Timing Verification Test or TVT, that selects the off-path input values to maximize the delay of the path. The tests produced by TVT result in path delays that are significantly higher than those obtained when off-path input values are selected arbitrarily by the ATPG if they are not mandated by the propagation conditions. TVT also considers different PVT corners that affect the selection of the longest paths. Experimental results for an industrial core show that TVT increases the path delays by up to 15.91% for a set of the longest paths needed to detect small cell-aware delay faults at different PVT corners.

The adverse effects of silent data errors (SDEs) on the operation of large data centers have been reported recently by hyper-scalar companies. SDEs tend to be elusive and are difficult to detect until they affect a particular application after the IC has been deployed in-field. Although the cause of SDEs ranges from manufacturing test escapes and design marginalities to design bugs, experimental data from the industry largely indicate that SDEs can be traced back to timing related issues that become more severe with aging and depend on the operating conditions of process, voltage and temperature (PVT). This paper describes a complete framework for monitoring the timing related issues under different operating conditions for test and functional operation. The framework has three components. The first component is a procedure for the identification of the longest paths that are prone to delay failures under different PVT corners. The second component is a programmable slack monitor design that monitors the changes in path delays within a detection window, and produces an alarm when a path is close to failure, with proximity to failure being a programmable feature. The third component is a procedure that determines the placement of the monitors in the design. Experimental results for an industrial design demonstrate the trade-offs related to the placement of monitors and the scenarios under which the monitors raise alarms.