Always-on electrocardiogram (ECG) monitoring is the best candidate to capture sporadic and intermittent irregular cardiac abnormalities. In this article, an ECG processor for always-on cardiac arrhythmia (CA) monitoring is presented. A delta quantization-based bitstream is used for the QRS morphology classification and detection of lethal cardiac arrhythmias like ventricular tachycardia (VT) and ventricular fibrillation (VF). Two tiny machine learning classifiers are used: A shallow ternary neural network classifier to perform QRS complex classification and a patient adaptive decision logic (ADL) approach for CA detection. The proposed processor utilizes only 100 Bytes of on-chip memory to store the necessary parameters and ECG data for the classifiers. The design of the ECG processor is based on high threshold standard cell 180-nm CMOS technology. The ASIC core occupies a die area measuring 0.49?mm2. In terms of power consumption, the measured total power is 65 nW, operating at a real-time clock frequency of 500 Hz and with a 1.4 V supply voltage. The performance of the proposed processor is evaluated on MIT-BIH Arrhythmia and CU Ventricular Tachyarrhythmia databases. It exhibits a high level of accuracy and is capable of detecting 12 types of CA with a sensitivity of?95.9%?and specificity of?99.71%.