In this paper, we present a method to miniaturize THz spectroscopic receivers into chip-scale systems by eliminating traditional spectrum analysis components such as wideband tunable THz sources, frequency multipliers, nonlinear mixing and amplifiers. This is achieved by extracting incident THz spectral signatures from the current distribution impressed on the surface of an on-chip antenna under the THz field incidence. A log-periodic tooth antenna is converted from a classical single port antenna into a massively multi-port structure with 84 detectors that sense the 2D current distribution. This method converts THz spectroscopy into a linear estimation problem and we apply robust estimation techniques such as LASSO, typically used in machine learning, to extract spectral signatures with noisy detectors with high sensitivity. The paper presents analytical and experimental results for the single chip operating at room temperature across 0.04-0.99 THz with 10 MHz accuracy in spectrum estimation of THz tones. The presented examples demonstrates that through a co-design approach of THz electromagnetics and electronics, such as by combining deep sub-wavelength near-field sensing and regression analyses, can enable a new class of THz chip-scale sensory systems.