Thermotolerant oscillatory neural networks enabled by electrical compensation

This work presents a thermotolerant oscillatory neural network (tONN) that is realized by using electrical compensation. The fundamental building block of the tONN consists of a pair of single-transistor oscillators (1T-Os), each connected to a load resistor (R_L), and a coupling capacitor (C_C) linking the two 1T-Os. Each 1T-O generates an oscillatory output voltage through the single-transistor latch (STL) phenomenon, whose latch-up voltage and oscillation frequency are inherently sensitive to temperature variations. The synchronization dynamics of a capacitively coupled pair of 1T-Os are investigated under dissimilar thermal conditions. A temperature imbalance between the paired 1T-Os degrades synchronization coherence as a result of a frequency mismatch. Normal tONN operation is achieved by controlling the gate voltage (V_G) of the 1T-Os. By electrically compensating for ambient temperature variations and restoring frequency matching, control over V_G enables coherent out-of-phase synchronization. The resulting thermal robustness is further demonstrated in edge detection and classification tasks. These findings demonstrate a device-level approach for realizing thermally robust ONN hardware using electrically compensated 1T-Os as compact, CMOS-compatible building blocks under anisothermal conditions.