Temperature mapping is essential in many biomedical studies and interventions to precisely control the tissue's thermal conditions for optimal treatment efficiency and minimal side effects. Based on the Grüneisen parameter's temperature dependence, photoacoustic (PA) imaging can provide relative temperature measurement, but it has been traditionally challenging to measure absolute temperatures without knowing the baseline temperature, especially in deep tissues with unknown optical and acoustic properties. Here, we report a new thermal-energy-memory-based photoacoustic thermometry (TEMPT). By illuminating the tissue with a burst of nanosecond laser pulses, TEMPT exploits the temperature dependence of the thermal energy lingering, which is probed by the corresponding PA signals acquired within the thermal confinement. A self-normalized ratiometric measurement cancels out temperature-irrelevant quantities and estimates the Grüneisen parameter. The temperature can then be evaluated, given the tissue's temperature-dependent Grüneisen parameter, mass density, and specific heat capacity. Unlike the conventional PA thermometry, TEMPT does not require the knowledge of tissue's baseline temperature, nor the optical properties. We have developed a mathematical model to describe the temperature dependence in TEMPT. We have demonstrated the feasibility of the temperature evaluation on tissue phantoms at 1.5 cm depth within a clinically relevant temperature range. Finally, as proof-of-concept, we applied TEMPT for temperature mapping during focused ultrasound treatment in mice in vivo at 2 mm depth. As a generic temperature mapping method, TEMPT is expected to find applications in thermotherapy of cancers on small animal models.