An associative memory device using a magnetic tunnel junction

Semiconductor technology for memory and logic devices has been developed to handle large amounts of data with a high operational speed. Attaining the functional abilities of the human brain is a primary goal for advanced modern electronics. Aside from neural networks with neurons and synapses as building blocks utilizing parallel processes to treat massive data, biological memory performs learning functions which are categorized as declarative (explicit) characteristics and non-declarative (implicit) characteristics [1]. While memory devices have a functional resemblance with the declarative memory because information is stored and recalled by declarative statements, artificial learning in neuromorphic approaches requires a distinct architecture differing from the standard CMOS-based electronic devices because the non-declarative memory recalls information by habituation, sensitization and classical conditioning [2]. In this work, we provide non-declarative memory using the specific switching characteristics [3] of an MTJ as the associative memory for the concept of classical conditioning in psychology which has been developed from Pavlov's observations describing conditioned responses by learning [4][6]. Our proposed device has a rather simple structure composed of a single MTJ for single association, and is capable of expanding association by adding a single MTJ for each event. The learning event is unquestionably determined without threshold fluctuation.