A life full of unconnected events, of errors that do not lead to any lessons and of emotions without the ability to remember them is no life at all. Memory is precisely the capacity that allows us to connect experiences, learn and make sense of our lives. In short, it allows us to build our story. The full range of this complex capacity’s neuroanatomical, neurobiological, neurophysiological, and psychological mechanism remain unknown and it presents a challenge for psychologists and neuroscientists who try to explain it. This review attempts to provide a rigorous overview that permits anyone who wants to approach the latest scientific findings on memory to do so, as well as to understand them and properly order them. We will focus on neuroanatomical, neurophysiological, and psychological mechanisms of the different types of memory.

Although knowledge of molecular mechanisms is important for constructing a complete vision of memory models, in this article we can only point out general traits as summarized in this introduction [for more information see (Kandel et al., 2014)]. In addition, knowledge gained from neuroimaging studies (Binder and Desai, 2011), as well as knowledge of the neural markers associated with memory (Meneses, 2015), will likely play a key role in future models of memory mechanisms, but in this review, as stated above, we focus mainly on neuroanatomical, neurophysiological, and psychological mechanisms.

We believe it is important to consider previous developments without which one cannot adequately understand the classifications of memories and the kinds of memory models that are now current in the scientific literature.

The three major classifications of memory that the scientific community deals with today are as follows: sensory memory, short-term memory, and long-term memory. Information from the world around us begins to be stored by sensory memory, making it possible for this information to be accessible in the future. Short-term memory refers to the information processed by the individual in a short period of time. Working memory performs this processing. Long-term memory allows us to store information for long periods of time. This information may be retrieved consciously (explicit memory) or unconsciously (implicit memory).

As Squire (2004) points out, the first theoretical approaches relevant to current neuroscience come from the 19th century. These include Maine de Biran (1804/1929) (Maine de Biran, 1929) who, at the beginning of the century, wrote of mechanical memory, sensitive memory, and representative memory. The philosopher James, and his book The Principles of Psychology (James, 1890), is also especially worth highlighting. Therein, James distinguishes between primary and secondary memory, thereby referring to short- and long-term memory, respectively.

The importance of Pavlov (1927) and Fitts and Posner (1967) are especially noteworthy during the first two thirds of the 20th century. Pavlov’s studies are related to a type of memory that later would be called associative memory. Meanwhile, Fitts and Posner’s studies are considered the first model to explain procedural memory.

Prior to the 60’s, most systematizations of memory distinguished a more mechanical type of memory related to the acquisition of skills, which is, in turn, related to activity of the intellect. Unlike what followed, debates in this period were mainly philosophical or based on psychological intuition (Ribot, 1881; Korsafoff, 1890).

Beginning in the 1960s, a series of experimental studies on how the brain stores information emerged, using animals and amnesic patients. Within this decade, Milner, Atkinson, and Shiffrin were especially important researchers.

The experimental modern era arguably began when Milner (1962) demonstrated, with HM experiments, that a seriously ill patient could acquire a new skill (hand-eye coordination) without any memory of having encountered the task before. “While this finding showed that memory is not unitary, discussions at the time tended to set aside motor skills as a special case representing a less cognitive form of memory. The suggestion was that the rest of memory is of one piece and is dependent on medial temporal lobe structures” (Squire and Wixted, 2016).

A few years later, Atkinson and Shiffrin (1968) proposed a modal model of memory that constitutes one of the most influential explanations for the existence of different components in the memory system. The importance of this model is such that it must be explained in the next section, but for now it should simply be mentioned that the modal model establishes the existence of short-term storage (ACP), which receives sensory information that is processed by sensory and data storehouses within long-term memory. This storage system can generate reasoning and new deductions from existing ones.

In the seventies, Tulving, Baddeley, and Hitch and Kandel’s investigations are especially noteworthy. Tulving (1972) first proposed the distinction between episodic memory and semantic memory. Baddeley and Hitch (1974) conducted research on the components of working memory. Both authors considered working memory as a limited capacity system that allows temporary storage and manipulation of information necessary to perform complex tasks such as understanding, learning, and reasoning. As explained later on, at first (1974), they proposed the existence of three subsystems within the multi-storehouse model of short-term memory: the central executive, a phonological or articulatory loop and a visuospatial sketchpad. Later, Baddeley (2000) included a fourth subsystem, the episodic buffer, which combines information from the subsystems in a form of temporal representation. Kandel (1976) proposed a model to explain the mechanism of operation in habituation and sensitization. To do this, he used the notion of non-associative memory, which, as we shall see, is one of the four types of non-declarative or implicit memory, like that which refers to new behaviors learned through repeated exposure to a single stimulus. According to Kandel, new behaviors can be classified into two processes: sensitization and habituation. On the one hand, for habituation, acetylcholine is progressively consumed, decreasing the effectiveness of the stimulus and thereby the motor response. On the other hand, the presence of serotonin in sensitization, secreted by another sensory nerve terminal, causes an excess of acetylcholine. An enhanced motor response thus emerges.

In the 1980s, the differences between declarative and non-declarative memories were consolidated and disseminated. This, together with contributions from Tulving and others, such as Di Lollo or Graf and Schacter, enabled a more precise classification of different types of memory. To date, Di Lollo’s model of iconic memory (Di Lollo, 1980) has been the most widely accepted and studied of the three existing types of sensory memory. As discussed in the next section, Di Lollo considered iconic memory a storage unit consisting of two components: the persistence of vision and information. Graf and Schacter (1985) proposed a general difference between declarative memory (explicit) and non-declarative memory (implicit/procedural). This stems from the distinction that Tulving (1972) proposed between the aforementioned episodic memory and semantic memory (both, as we will see, are currently included in declarative memory).

In the 90’s, a classification of the types of memory emerged, but the way they act and their interrelation was still unclear. In order to clarify its operation, Packard and McGaugh (1996) proposed that memory systems operate independently and in parallel. For example, an adverse event in childhood (e.g., seeing your grandfather being run over by a combine) can, on the one hand, consolidate as a stable declarative memory for the event itself (the sound of a combine always makes you remember that moment-episodic memory) and, on the other hand, can crystallize in non-declarative memory and result in a phobia experienced as a personality trait rather than as a mere memory (being near a combine will always produce panic and induces a desire to escape that situation-associative memory). Several authors (Tulving et al., 1982) had already mentioned the idea of priming as a separate type of memory, but it was not until the 90’s that experiments were conducted to show it (Hamann and Squire, 1997; Stark and Squire, 2000; Levy et al., 2004). These studies show that severely amnesic patients can exhibit completely intact priming while performing memory tests that include conventional recognition of the same test items (Squire, 2004).

Thanks to the development of new 21st century technologies, researchers have been able to more accurately locate brain areas that are associated with different types of memory. Although this pertains to topics to be addressed in detail in the next section, there are two examples that we consider significant to the application of these new techniques and the significant progress made in understanding memory storage. On the one hand, Ergorul and Eichenbaum’s experiment (Ergorul and Eichenbaum, 2004) shows that animals are able to remember the “context in which they experienced specific stimuli, and that this capacity also depends on the hippocampus” (Dickerson and Eichenbaum, 2010). This process is closely related to the formation of episodic memory. On the other hand, neuroimaging studies that Binder and Desai (2011) conducted show “two striking results: the participation of modality-specific sensory, motor, and emotion systems in language comprehension, and the existence of large brain regions that participate in comprehension tasks but are not modality-specific.” With this in mind, Binder and Desai (2011) claims that semantic memory consists of two representations, including a specific mode and a supramodal mode. Again, this will be explained in more detail in what follows.

The research of the cellular and molecular substrates of memory has received much attention since Lomo (1966) described in the 60’s “a cellular model of experience-dependent plasticity—long-term potentiation (LTP)” (Kandel et al., 2014). According to Lisman et al. (2012): “LTP is a process whereby brief periods of synaptic activity can produce a long-lasting increase in the strength of a synapse, as shown by an increase in the size of the excitatory postsynaptic current.” NMDA receptors are double-gated, as their activation requires both postsynaptic membrane depolarization as well as presynaptic release of glutamate. Once activated by these conditions, NMDA receptors trigger a strong postsynaptic influx of Ca2+ that induce LTP through a variety of pathways including CaMKII, PKC, PKA, and MAPK (Kandel et al., 2014).

With this brief historical and conceptual introduction laid out, we intend to delve into different types of memory in order to present the models that the scientific community has most accepted thus far. In the last section, and before the glossary, we identify the likely directions for future research. Now we turn on to our main task, presenting an overview of the latest scientific findings on memory, classified according to different types and mechanisms.

“Sensory memory is the capacity for briefly retaining the large amounts of information that people encounter daily” (Siegler and Alibali, 2005). There are three types of sensory memory: echoic memory, iconic memory, and haptic memory. Iconic memory retains information that is gathered through sight, echoic memory retains information gathered through auditory stimuli and haptic memory retains data acquired through touch.

Scientific research has focused mainly on iconic memory; information on echoic and haptic memory is comparatively scarce. Thus, taking into account the goals of this article and that it is aimed at a higher education audience, presenting iconic memory as a paradigm of sensory memory is sufficient for an introductory overview.

Iconic memory retains information from the sense of sight with an approximate duration of 1 s. This reservoir of information then passes to short-term vision memory (which is analogous, as we shall see shortly, to the visuospatial sketchpad with which working memory operates).

Di Lollo’s model (Di Lollo, 1980) is the most widely accepted model of iconic memory. Therein, he considered iconic memory a storehouse constituted by two components: the persistence of vision and information.

(a) Persistence of vision. Iconic memory corresponds to the pre-categorical representation image/visual that remains between 100 and 300 ms. It is sensitive to physical parameters, such that it depends on retinal photoreceptors (rods and cones). It also depends on various cells in the visual system and on retinal ganglion cells M (transition cells) and P (sustained cells). It concludes its representation in the primary visual cortex (V1) of the occipital lobes. “The occipital lobe is responsible for processing visual information” (Kamel and Malik, 2014).

(b) Persistence of information. Iconic memory is a storehouse of information that lasts 800 ms and that represents a codified and already categorized version of the visual image. It plays the role of storehouse for post-categorical memory, which provides visual short-term memory with information to be consolidated. For this, it travels through the ventral route (V) (V1 → V2 → V5 → inferior temporal cortex).

Subsequent research on visual persistence from Coltheart (Coltheart, 1983) and Sperling’s studies (Sperling, 1960) on the persistence of information led to the definition of three characteristics pertaining to iconic memory: a large capacity, a short duration, and a pre-categorical nature.

Sperling (1960) demonstrated this large capacity after presenting the results of his total and partial reports. The full report consisted in presenting a 3 × 3 or 3 × 4 matrix of alphanumeric characters for a short period of time to subjects and later asking them which characters they remembered. On the other hand, in the partial report, subjects were directed to remember the characters in a row specifically assigned to them in the instructions. The total report’s results showed that subjects were only able to recall between 3 or 4 letters of the total number. However, in the partial report, subjects remembered around 75% of those that were asked. In extrapolating the partial report’s data to the total, it follows that individuals could report 9 of the 12 letters contained in the instructions (80% of the total), thus demonstrating a large capacity.

Regarding short-term, Sperling interpreted the results of the partial report as due to the rapid decline of the visual sign and reaffirmed this short duration by obtaining a decrease in the number of letters reported by the subject in delaying the audio signal for choosing a row to remember in the presentation. Averbach and Coriell’s experiments (Averbach and Coriell, 1961) corroborated Sperling’s conclusion; they presented a variety of letters for a certain period of time to the subject. After each letter, and in the same position, they showed a particular visual sign. The participant’s task was to name the letter that occupied the position of the visual sign. When the visual sign appeared immediately after the letters, participants could correctly name the letter that occupied the position of the sign, however, as the presentation of the sign became more delayed, participant performance worsened. These results also show the rapid decline of visual information.

Finally, regarding its pre-categorical nature, Sperling considered the information contained in this storehouse as physical information that maintains the raw data that is not related to the meaning of stimuli. Subsequently, evidence has been obtained that this system is not entirely pre-categorical (Loftus et al., 1992) since the task improves when the stimuli to remember are letters or numbers instead of meaningless forms.

Short-term memory is the ability to keep a small amount of information available for a short period of time. Atkinson and Shiffrin’s modal model (Atkinson and Shiffrin, 1968) is one of the most influential explanations for the existence of different components in the memory system (Figure 1). This model has some similarities with Broadbent’s previous model (Broadbent, 1958). The modal model establishes the existence of a short-term storehouse with limited capacity. The short-term storehouse receives sensory information processed by sensory storehouses and data in long-term memory. In addition, the short-term storehouse can also send information to the structures involved in long-term memory. This storehouse can generate reasoning and new deductions from existing ones. This model implies that the short-term storehouse functions as a kind of working memory, a system to retain and manipulate information temporarily as part of a wide range of essential cognitive tasks such as learning, reasoning, and understanding. They, in turn, give short-term storage central importance in the overall processing of information by attributing to it the role of controlling the executive system, responsible for the coordination and control of many complex subroutines in charge of acquiring new material and recovering old material in long-term storage.

Despite the explanatory power of Atkinson and Shiffrin’s model, there were a number of issues that this model could not resolve, causing criticism of it. For example, this model implies that the longer an item remains in memory, the more likely it is to be transferred to long-term storage. But studies like those of Tulving and Pearlstone (1966) and Craik and Watkins (1973) show that said relationship does not exist.

Given these criticisms, new models began to appear to explain memory, such as those from Cowan (1988, 1995, 1999) and Goldman-Rakic (1995). Among them, Craik and Lockhart’s process model (Craik and Lockhart, 1972) and Baddeley and Hitch’s structural model (Baddeley and Hitch, 1974) were the most prominent; the latter is the most commonly accepted one today and thus we will focus on it in this article.

As an introduction, it can be argued that Craik and Lockhart (1972) understood memory not as a process through which information is deepened at higher levels until it becomes part of long-term memory, but rather as a system of storehouses. Despite an emphasis on information processing (instead of structure), they continued to accept the existence of short-term memory as independent from long-term memory. For their part, Baddeley and Hitch’s proposal (Baddeley and Hitch, 1974) contemplated a multi-component working memory instead of a storage unit in the short term.

Long-term memory refers to unlimited storage information to be maintained for long periods, even for life. There are two types of long-term memory: declarative or explicit memory and non-declarative or implicit memory.

Explicit memory refers to information that can be consciously evoked. There are two types of declarative memory: episodic memory and semantic memory. For its part, implicit memory encompasses all unconscious memories, such as certain abilities or skills. There are four types of implicit memory, including procedural, associative, non-associative, and priming.

In spite of recent progress, a number of important questions remain to be tackled.

Many of these questions have to do with molecular processes of memory consolidation, retrieval, and decay. Take, for instance, the processes underlying the LTP of synaptic strength among neurons of the hippocampus (Dong et al., 2015). In their review, Hardt et al. (2013) point out that while the “molecular processes involved in establishing LTP have been characterized well, the decay of early and late LTP is poorly understood.” One possibility that has recently been suggested is that LTP decay is mediated by AMPAR endocytosis, which in turn implies that inhibition of this process could preserve LTP and help to prevent memory loss (Dong et al., 2015).

Other recent work shows the critical role of dopamine as a signal that promotes the stable incorporation of novel information into long-term hippocampal memory (Otmakhova et al., 2013). Indeed, dopamine neurons can be activated by novelty in the absence of reward and it is thought that this activation occurs via a polysynaptic pathway that runs from the hippocampus to the dopamine cells of the VTA. However, many aspects of this process remain unclear (see Otmakhova et al., 2013).

Also requiring further investigation are the molecular processes involved in the regulation of protein synthesis related to memory. It is now thought that protein synthesis is not only involved in the consolidation of new memories, but must also be used to re-consolidate memories that have been degraded or “destabilized” as a result of retrieval. In a recent article, Jarome et al. (2016) indicate that CaMKII controls the re-consolidation process through the regulation of proteasome activity. Another mechanism for the regulation of protein synthesis involves MicroRNAs (miRNAs), a class of short, non-coding RNAs. By regulating components of pathways required for learning and memory, miRNAs modulate the influence of epigenetics on cognition in the normal and diseased brain (Saab and Mansuy, 2014).

Another set of important questions relates to the long-standing hypothesis of “Hebbian learning”—the strengthening of synapses between neurons with correlated activity—and its role in memory. At the same time that we are learning more about the mechanisms involved in Hebbian plasticity, we are also learning about how these mechanisms are complemented by synaptogenesis and neuromodulatory processes. Recent research has shown that synaptogenesis is not only important during development, but also plays a central role in associative learning and memory. Synaptogenesis can be triggered by neuron–astrocyte or neuron–neuron contact, and mediated by cell-adhesion proteins including neurexin/neuroligin, Eph receptors, and cadherins, which activate intracellular signaling pathways involving cofilin, GTPases, and other proteins (for a review see Nelson and Alkon, 2015). Others have proposed that Hebbian processes, while important, are not sufficient for memory formation, and must be supported by the activation of neuromodulatory processes, especially in the case of associative aversive learning (Johansen et al., 2014). Another study found evidence for both Hebbian and anti-Hebbian mechanisms of synaptic plasticity, indicating that the mechanisms of learning are highly adaptable (Koch et al., 2013).

Finally, other questions relate to the special role of the amygdala in the emotional enhancement of memory consolidation. It has long been known that emotional arousal contributes to the selection and consolidation of memory. It has also been shown that previously weak or inconsequential information can be strengthened retroactively through an emotional learning experience (Dunsmoor et al., 2015). Evidence suggests that it is the amygdala that is most responsible for the enhancement of memory (de Voogd et al., 2016). As suggested by a review, these developments “will likely lead to an updated view of the amygdala as a critical nexus within large-scale networks supporting different aspects of memory processing for emotionally arousing experiences” (Hermans et al., 2014). Moreover, this research is likely to have important implications for the treatment of psychological disorders (Beckers and Kindt, 2017).

There are three main forms of memory: sensory memory, short-term memory, and long-term memory (Figure 3). Sensory memory refers to the retention of information coming from the senses. Short-term memory refers to information processed in a short period of time. Working memory performs this processing. Working memory consists of four elements that process information: the central executive (attention control), the visuospatial sketchpad (creates and maintains a visuospatial representation), the phonological buffer (stores and consolidates new words), and the episodic buffer (stores and integrates information from different sources). Long-term memory allows us to store information for long periods of time. This information may be retrieved consciously (explicit memory) or unconsciously (implicit memory). Explicit memory consists of episodic memory (time-related events) and semantic memory (concepts and meanings). Implicit memory has, in turn, procedural memory (motor and executive skills), associative memory (classical and operant conditioning), non-associative memory (sensitization and habituation), and priming (a primary stimulus influencing a secondary one).

Finally, the following glossary includes commentary about the terminology that, in our opinion, is essential for an introductory overview, enabling interested students and professionals to effectively approach the latest memory-related discoveries. This commentary is not intended as an exhaustive definition, but rather collects relevant information to situate the reader within a complex panorama.

Associative memory: refers to the storage and retrieval of information resulting from an association (i.e., resulting from an association with other information). Two types of conditioning are involved in its acquisition: classical conditioning and operant conditioning. Classical conditioning is a kind of associative learning between stimuli and behavior, and operant conditioning is a form of learning in which new behaviors develop in terms of their consequences.

Conceptual short-term memory/episodic buffer: This is a temporary storage system capable of integrating information from different sources that is probably controlled by the central executive. It is episodic in that it has episodes in which information is integrated through space and, potentially, extended through time.

Echoic memory: sensory memory that receives and processes auditory information.

Episodic memory: “involves the ability to learn, store, and retrieve information about unique personal experiences that occur in daily life. These memories typically include information about the time and place of an event, as well as detailed information about the event itself” (Dickerson and Eichenbaum, 2010).

Explicit/declarative memory: refers to conscious memories of previously stored experiences, facts and concepts that are verifiable through a verbal reporting of them (Tulving, 1972).

Haptic memory: sensory memory that receives and processes information from the sense of touch.

Iconic memory: visual-sensory memory that receives and processes visual stimuli.

Implicit/non-declarative memory: this encompasses all unconscious memories, as well as certain abilities or skills. There are four types of implicit memory: procedural, associative, non-associative, and priming memory.

Long-term memory: “refers to the unlimited, continuing memory store that can hold information over lengthy periods of time, even for an entire lifetime. Long-term memory is mainly preconscious and unconscious. Information in long-term memory is to a great extent outside of our awareness, but can be called into working memory to be used when needed. Some of this information is easy to recall, but some is much more difficult to access” (Brodziak et al., 2013).

Non-associative memory: refers to newly learned behavior due to repeated exposure to a single stimulus. The new behavior can be classified into two processes: sensitization and habituation.

Perceptual memory: memory acquired through the senses. It includes a lot of individual experience; it ranges from the simplest forms of sensory memory to the most abstract knowledge.

Priming: an effect whereby exposure to certain stimuli influences the response to subsequently presented stimuli.

Procedural memory: a memory area involved in remembering executive and motor skills necessary to perform a task. It is an executive system that guides activity and usually works on an unconscious level. When necessary, procedural memories are automatically retrieved for use in the implementation of integrated procedures related to motor and intellectual skills.

Semantic memory: refers to the memory of meanings, interpretations and concepts related to facts, information and general knowledge about the world. Semantic memory gives meaning to words and phrases that would otherwise be meaningless and allows for learning based on past experience (Kolb and Whishaw, 2003).

Sensory memory: “Sensory memory is the capacity for briefly retaining the large amounts of information that people encounter daily” (Siegler and Alibali, 2005).

Short-term memory: is the ability to keep a small amount of information available for a short period of time. “Short-term memory should be distinguished from working memory, which refers to structures and processes used for temporarily storing and manipulating information. The relationship between short-term memory and working memory is presented variously by different theories. The notion of working memory is broader and more general because it refers to structures and processes used for temporarily stored and manipulated information” (Brodziak et al., 2013).

Working memory: “The term working memory refers to a brain system that provides temporary storage and manipulation of the information necessary for such complex cognitive tasks as language comprehension, learning and reasoning” (Baddeley, 1992).

Visual memory: constituted by iconic memory, visual short-term and long-term memory.

Visual short-term memory/visuospatial sketchpad: sketchpad’s main function is to create and maintain a visuospatial representation that persists through the irregular form found in eye movement and that characterizes our exploration of the visual world (Luck, 2007).

FG drafting of sensory and short-term memory; bibliographical review; introduction and conclusion. Revising critically and final approval of the version to be published. EC drafting of long-term memory (declarative and non-declarative); bibliographical review.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.