Along with the popularization and rapid development of cloud-computing, more and more individuals and enterprises choose to store their data in cloud servers. However, in order to protect data privacy and deter illegal accesses, the data owner has to encrypt his data before outsourcing it to the cloud server. In this situation, searchable encryption, especially searchable symmetric encryption (SSE) has become one of the most important techniques in cloud-computing area. In the last few years, researchers have presented many secure and efficient SSE schemes. Like traditional encryption, the security of all existing SSE schemes are based on the assumption that the data owner holds a secret key that is unknown to the adversary. Unfortunately, in practice, attackers are often able to obtain some or even all of the data owner's secret keys by a great variety of inexpensive and fast side channel attacks. Facing such attacks, all existing SSE schemes are no longer secure. In this paper, we investigate how to construct secure SSE schemes with the presence of memory attack. We firstly propose the formal definition of memory leakage-resilient searchable symmetric encryption (MLR-SSE, for short). Based on that, we present one adaptive MLR-SSE scheme and one efficient non-adaptive dynamic MLR-SSE scheme based on physical unclonable functions (PUFs), and formally prove their security in terms of our security definitions. Two symmetric searchable encryption protocols against memory leakage are proposed.The mechanism mainly relies on the character of physical unclonable functions.The attacker considered here is a non-volatile memory attacker.