Associative Memory Cells Research Articles

Piriform cortical glutamatergic and GABAergic neurons express coordinated plasticity for whisker-induced odor recall.

Neural plasticity occurs in learning and memory. Coordinated plasticity at glutamatergic and GABAergic neurons during memory formation remains elusive, which we investigate in a mouse model of associative learning by cellular imaging and electrophysiology. Paired odor and whisker stimulations lead to whisker-induced olfaction response. In mice that express this cross-modal memory, the neurons in the piriform cortex are recruited to encode newly acquired whisker signal alongside innate odor signal, and their response patterns to these associated signals are different. There are emerged synaptic innervations from barrel cortical neurons to piriform cortical neurons from these mice. These results indicate the recruitment of associative memory cells in the piriform cortex after associative memory. In terms of the structural and functional plasticity at these associative memory cells in the piriform cortex, glutamatergic neurons and synapses are upregulated, GABAergic neurons and synapses are downregulated as well as their mutual innervations are refined in the coordinated manner. Therefore, the associated activations of sensory cortices triggered by their input signals induce the formation of their mutual synapse innervations, the recruitment of associative memory cells and the coordinated plasticity between the GABAergic and glutamatergic neurons, which work for associative memory cells to encode cross-modal associated signals in their integration, associative storage and distinguishable retrieval.

Coordinated Plasticity among Glutamatergic and GABAergic Neurons and Synapses in the Barrel Cortex Is Correlated to Learning Efficiency.

Functional plasticity at cortical synapses and neurons is presumably associated with learning and memory. Additionally, coordinated refinement between glutamatergic and GABAergic neurons occurs in associative memory. If these assumptions are present, neuronal plasticity strength and learning efficiency should be correlated. We have examined whether neuronal plasticity strength and learning efficiency are quantitatively correlated in a mouse model of associative learning. Paired whisker and odor stimulations in mice induce odorant-induced whisker motions. The fully establishment of this associative memory appears fast and slow, which are termed as high learning efficiency and low learning efficiency, respectively. In the study of cellular mechanisms underlying this differential learning efficiency, we have compared the strength of neuronal plasticity in the barrel cortices that store associative signals from the mice with high vs. low learning efficiencies. Our results indicate that the levels of learning efficiency are linearly correlated with the upregulated strengths of excitatory synaptic transmission on glutamatergic neurons and their excitability, as well as the downregulated strengths of GABAergic neurons' excitability, their excitatory synaptic inputs and inhibitory synaptic outputs in layers II~III of barrel cortices. The correlations between learning efficiency in associative memory formation and coordinated plasticity at cortical glutamatergic and GABAergic neurons support the notion that the plasticity of associative memory cells is a basis for memory strength.

Associative Memory Extinction Is Accompanied by Decayed Plasticity at Motor Cortical Neurons and Persistent Plasticity at Sensory Cortical Neurons.

Associative memory is essential for cognition, in which associative memory cells and their plasticity presumably play important roles. The mechanism underlying associative memory extinction vs. maintenance remains unclear, which we have studied in a mouse model of cross-modal associative learning. Paired whisker and olfaction stimulations lead to a full establishment of odorant-induced whisker motion in training day 10, which almost disappears if paired stimulations are not given in a week, and then recovers after paired stimulation for an additional day. In mice that show associative memory, extinction and recovery, we have analyzed the dynamical plasticity of glutamatergic neurons in layers II–III of the barrel cortex and layers IV–V of the motor cortex. Compared with control mice, the rate of evoked spikes as well as the amplitude and frequency of excitatory postsynaptic currents increase, whereas the amplitude and frequency of inhibitory postsynaptic currents (IPSC) decrease at training day 10 in associative memory mice. Without paired training for a week, these plastic changes are persistent in the barrel cortex and decayed in the motor cortex. If paired training is given for an additional day to revoke associative memory, neuronal plasticity recovers in the motor cortex. Our study indicates persistent neuronal plasticity in the barrel cortex for cross-modal memory maintenance as well as the dynamical change of neuronal plasticity in the motor cortex for memory retrieval and extinction. In other words, the sensory cortices are essential for long-term memory while the behavior-related cortices with the inability of memory retrieval are correlated to memory extinction.

AM06: the Associative Memory chip for the Fast TracKer in the upgraded ATLAS detector

This paper describes the AM06 chip, which is a highly parallel processor for pattern recognition in the ATLAS high energy physics experiment. The AM06 contains memory banks that store data organized in 18 bit words; a group of 8 words is called “pattern”. Each AM06 chip can store up to 131 072 patterns. The AM06 is a large chip, designed in 65 nm CMOS, and it combines full-custom memory arrays, standard logic cells and serializer/deserializer IP blocks at 2 Gbit/s for input/output communication. The overall silicon area is 168 mm2 and the chip contains about 421 million transistors. The AM06 receives the detector data for each event accepted by Level-1 trigger, up to 100 kHz, and it performs a track reconstruction based on hit information from channels of the ATLAS silicon detectors. Thanks to the design of a new associative memory cell and to the layout optimization, the AM06 consumption is only about 1 fJ/bit per comparison. The AM06 has been fabricated and successfully tested with a dedicated test system.

Associative memory cells: Formation, function and perspective

Associative learning and memory are common activities in life, and their cellular infrastructures constitute the basis of cognitive processes. Although neuronal plasticity emerges after memory formation, basic units and their working principles for the storage and retrieval of associated signals remain to be revealed. Current reports indicate that associative memory cells, through their mutual synapse innervations among the co-activated sensory cortices, are recruited to fulfill the integration, storage and retrieval of multiple associated signals, and serve associative thinking and logical reasoning. In this review, we aim to summarize associative memory cells in their formation, features and functional impacts.

Associative memory cells: Formation, function and perspective.

Associative learning and memory are common activities in life, and their cellular infrastructures constitute the basis of cognitive processes. Although neuronal plasticity emerges after memory formation, basic units and their working principles for the storage and retrieval of associated signals remain to be revealed. Current reports indicate that associative memory cells, through their mutual synapse innervations among the co-activated sensory cortices, are recruited to fulfill the integration, storage and retrieval of multiple associated signals, and serve associative thinking and logical reasoning. In this review, we aim to summarize associative memory cells in their formation, features and functional impacts.

Associative Memory Cells are Recruited to Encode Triple Sensory Signals via Synapse Formation

Associative learning is a common form of information acquisition, and associative memory is essential for logical reasoning and associative thinking. In addition to memory cells for two associated signals, here we report the recruitment of associative memory cells that encode triple sensory signals. Paired mouse whisker, olfaction and tail stimulations lead to odorant-induced and tail-induced whisker motions. In the mice of showing cross-modal reflexes, barrel cortical neurons and astrocytes become to encode odor and tail signals alongside whisker signal. Such neurons receive the new synapse innervations from the axons of the piriform and S1-tail cortical neurons, in addition to the innate synapse innervations from the axons of the thalamic neurons. The formation of new synapse innervations and the recruitment of associative memory cells require the upregulations of miRNA-324-5p and miRNA-133a-3p that downregulate tau-tubulin kinase-1 (Ttbk1) and methylcytosine dioxygenase Tet3. The associated activations of the sensory cortices elicit their mutual synapse innervations that recruit memory cells to store triple associated signals via epigenetic processes. Furthermore, after the extinction of cross-modal reflexes, the refinements of the synapses and memory cells in the barrel cortex are well maintained, while plasticity in the M1-motor cortex decreases. These results indicate that the acquired signals remain stored in the sensory cortices and the decay of memory retrieval is due to the decreased refinement of their downstream cortical regions in neural circuits from the sensory cortices to the memory-presentation cortices. In terms of the significance of our discoveries, the storages of multiple signals in individual neurons may expand memory volumes, strengthen cognition capabilities and facilitate creative inspiration productions. Information storages in the sensory cortices confer them to signify the sources of the retrieved memory signals. This study is supported by National Basic Research Program (2013CB531304 and 2016YFC1307100) and Natural Science Foundation China (81671071 and 81471123) to JHW.

Associations of Unilateral Whisker and Olfactory Signals Induce Synapse Formation and Memory Cell Recruitment in Bilateral Barrel Cortices: Cellular Mechanism for Unilateral Training Toward Bilateral Memory.

Somatosensory signals and operative skills learned by unilateral limbs can be retrieved bilaterally. In terms of cellular mechanism underlying this unilateral learning toward bilateral memory, we hypothesized that associative memory cells in bilateral cortices and synapse innervations between them were produced. In the examination of this hypothesis, we have observed that paired unilateral whisker and odor stimulations led to odorant-induced whisker motions in bilateral sides, which were attenuated by inhibiting the activity of barrel cortices. In the mice that showed bilateral cross-modal responses, the neurons in both sides of barrel cortices became to encode this new odor signal alongside the innate whisker signal. Axon projections and synapse formations from the barrel cortex, which was co-activated with the piriform cortex, toward its contralateral barrel cortex (CBC) were upregulated. Glutamatergic synaptic transmission in bilateral barrel cortices was upregulated and GABAergic synaptic transmission was downregulated. The associative activations of the sensory cortices facilitate new axon projection, glutamatergic synapse formation and GABAergic synapse downregulation, which drive the neurons to be recruited as associative memory cells in the bilateral cortices. Our data reveal the productions of associative memory cells and synapse innervations in bilateral sensory cortices for unilateral training toward bilateral memory.

Both Glutamatergic and Gabaergic Neurons are Recruited to be Associative Memory Cells

The recruitment of associative memory cells and their working principle for signal storage were studied by in vivo two-photon calcium imaging, AAV-tagged neural tracing, electrophysiology and microRNA analyses. Paired whisker and odor stimulations led to reciprocal cross-modal reflexes, odorant-induced whisker motion and whisker-induced olfaction response. In the mice of expressing cross-modal memory, the piriform and barrel cortices mutually innervated through their axon projections and the new synapses from these axons to target cells. Glutamatergic and GABAergic neurons in the barrel cortex became to encode the newly acquired odor signal alongside innate whisker signal. Piriform cortical neurons encoded the learnt whisker signal alongside innate odor signal. These associative memory neurons were able to distinguish new signal and innate signal as well as to tell their historical association. These associative memory neurons received new synaptic inputs along with native synaptic inputs. In addition, the newly learnt signals were sent to the contralateral cortices through new axons and synapses for unilateral learning toward bilateral memory. The antagomirs of microRNA-324p and microRNA-133a reduced the formation of cross-modal memory, the recruitment of associative memory cells and the formation of new synapses. The co-activations of the sensory cortices initiate their mutual synaptic innervations, which recruit their target cells to be associative memory cells. Our study reveals associative memory cells based on these criteria. They encode associative signals (two or more), receive the synaptic inputs from signals’ origins and send the output signals to control behaviors. The recruitment of associative memory cells is downregulated by changing gene expression, or vice versa. Associative memory cells are recruited in the sensory cortices from our model, compared to fear learning that induces extensive associations between the whole brain excited by electrical shock and the auditory cortex by sound.

Coordinated Plasticity between Barrel Cortical Glutamatergic and GABAergic Neurons during Associative Memory.

Neural plasticity is associated with memory formation. The coordinated refinement and interaction between cortical glutamatergic and GABAergic neurons remain elusive in associative memory, which we examine in a mouse model of associative learning. In the mice that show odorant-induced whisker motion after pairing whisker and odor stimulations, the barrel cortical glutamatergic and GABAergic neurons are recruited to encode the newly learnt odor signal alongside the innate whisker signal. These glutamatergic neurons are functionally upregulated, and GABAergic neurons are refined in a homeostatic manner. The mutual innervations between these glutamatergic and GABAergic neurons are upregulated. The analyses by high throughput sequencing show that certain microRNAs related to regulating synapses and neurons are involved in this cross-modal reflex. Thus, the coactivation of the sensory cortices through epigenetic processes recruits their glutamatergic and GABAergic neurons to be the associative memory cells as well as drive their coordinated refinements toward the optimal state for the storage of the associated signals.

Efficient Associative Computation with Discrete Synapses.

Neural associative networks are a promising computational paradigm for both modeling neural circuits of the brain and implementing associative memory and Hebbian cell assemblies in parallel VLSI or nanoscale hardware. Previous work has extensively investigated synaptic learning in linear models of the Hopfield type and simple nonlinear models of the Steinbuch/Willshaw type. Optimized Hopfield networks of size n can store a large number of about n(2)/k memories of size k (or associations between them) but require real-valued synapses, which are expensive to implement and can store at most C = 0.72 bits per synapse. Willshaw networks can store a much smaller number of about n(2)/k(2) memories but get along with much cheaper binary synapses. Here I present a learning model employing synapses with discrete synaptic weights. For optimal discretization parameters, this model can store, up to a factor ζ close to one, the same number of memories as for optimized Hopfield-type learning--for example, ζ = 0.64 for binary synapses, ζ = 0.88 for 2 bit (four-state) synapses, ζ = 0.96 for 3 bit (8-state) synapses, and ζ > 0.99 for 4 bit (16-state) synapses. The model also provides the theoretical framework to determine optimal discretization parameters for computer implementations or brainlike parallel hardware including structural plasticity. In particular, as recently shown for the Willshaw network, it is possible to store C(I) = 1 bit per computer bit and up to C(S) = log n bits per nonsilent synapse, whereas the absolute number of stored memories can be much larger than for the Willshaw model.

Neurons in the barrel cortex turn into processing whisker and odor signals: a cellular mechanism for the storage and retrieval of associative signals

Associative learning and memory are essential to logical thinking and cognition. How the neurons are recruited as associative memory cells to encode multiple input signals for their associated storage and distinguishable retrieval remains unclear. We studied this issue in the barrel cortex by in vivo two-photon calcium imaging, electrophysiology, and neural tracing in our mouse model that the simultaneous whisker and olfaction stimulations led to odorant-induced whisker motion. After this cross-modal reflex arose, the barrel and piriform cortices connected. More than 40% of barrel cortical neurons became to encode odor signal alongside whisker signal. Some of these neurons expressed distinct activity patterns in response to acquired odor signal and innate whisker signal, and others encoded similar pattern in response to these signals. In the meantime, certain barrel cortical astrocytes encoded odorant and whisker signals. After associative learning, the neurons and astrocytes in the sensory cortices are able to store the newly learnt signal (cross-modal memory) besides the innate signal (native-modal memory). Such associative memory cells distinguish the differences of these signals by programming different codes and signify the historical associations of these signals by similar codes in information retrievals.

Inexact match associative memory cell

A new associative memory cell is described and analysed using SPICE3d1. The CMOS cell uses current summation to compute, in parallel, Hamming distances between the search key and each word in the memory. For 32 bit words, SPICE simulations of a 2μm process show a delay of 4ns/bit for Hamming distances less than three.

A two-dimensional, distributed logic architecture

The authors present a novel, very fine grain associative architecture. This architecture maintains both a high degree of flexibility and fine graininess. This is done by reducing each processor to an associative memory cell. Unlike other associative memory processors, this architecture uses a two-dimensional interconnect and a physically compact memory structure. Arithmetic operations are based on the use of a redundant number system. These features provide a high level of performance. This is particularly true for certain two-dimensional problems which can be solved very efficiently on the proposed architecture.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>

A Josephson ternary associative memory cell

The authors describe a three-valued content-addressable memory cell using a Josephson complementary ternary logic (JCTL) circuit. The memory cell can perform the operations of searching, writing and reading in the ternary logic system. The principle of the memory circuit is illustrated in detail by using the threshold characteristics of the JCTL. Computer simulations were performed to investigate how high-performance operation can be achieved. Simulation results show that the cycle time of memory operation is 120 ps, power consumption is about 0.5 mu W/cell, and tolerances of writing and reading operation are +or-15% and +or-24% respectively.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>

Static analysis of a resistor‐load‐type D2 CMOS (dual depletion CMOS) associative memory cell by using the state diagram method

AbstractIn addition to functions required for conventional RAM cells, an associative memory cell requires a function for interrogating its input information and memory information. These requirements increase the number of components per bit, and create the problem of increased cell area when the components are integrated. This paper proposes a new associative memory cell using dual depletion CMOS (D2 CMOS) devices which have fewer components per bit than conventional associative memory cells so that they are more suitable for an integration. By using “the state diagram method,” the static characteristics of the cell were analyzed. The conditional equations of the load resistance which makes the cell functional as an associative memory were obtained and the memory characteristics examined. The writing characteristics involving the conditions for the shift of a state and the writing voltage also were analyzed. In addition, the reading characteristics and the interrogating characteristics were examined. The results of the analyses agree fairly well with the results of the simulation using the circuit analysis programme SPICE 2.

An 8-kbit content-addressable and reentrant memory

A 256-word/spl times/32-bit associated memory, referred to as the Content Addressable and Reentrant Memory (CARM), with a 100-ns cycle time is described. The high bit density of the device is realized by a small-size associative memory cell (30/spl times/36 /spl mu/m/SUP 2/) with 2-/spl mu/m CMOS technology, while a double-layer metallization technique, new circuits for the control-signal propagation, and a hierarchical structure for the address encoder of the chip allow fast access. This device has reentrant mode operation, where the on-chip garbage collection or data storage is accomplished conditionally. One of the practical applications of this device, a high-speed matching unit for dataflow computers, is also discussed.

A content-addressable memory cell with MNOS transistors

Describes a new associative memory cell in which MNOS transistors are used as storage elements. The memory can perform functions as a read-only memory and at the same time as a read-write memory. The cell can be read as a random-access memory or as a content-addressable memory. As a CAM certain bits can be masked out, i.e., not compared with the stored bits. The comparison can also be controlled from the memory by the stored words. Since the word length or combinations of normal words can be stored in one word of the memory, fewer memory cells are needed than in an ordinary memory. Searches for groups of words (prime implicands) can be performed. Memory cells with an area of 5000-m- have been built to demonstrate the feasibility of the MNOS-CAM.

Low-cost associative memory

The authors describe the design of two high-density MOS associative memory cells. The first cell is suitable for data management applications, having three internal states including the DON'T CARE (mask) condition. The second cell is suitable for parallel processing applications and has the capability of selected bit writing. Both cells consume about 20 mil/SUP 2/ of silicon area, which allows the implementation of 512 bits on a chip. Varactor bootstrapping is used to enhance the storage node voltage for improved cell operation. The use of a compatible bipolar transistor is discussed.

Design for a high-speed m.o.s. associative memory

An experimental 64-bit m.o.s. associative memory has been developed from a limit-case design study. Speeds in excess of 50 MHz are reported at a cost per bit that could approach eight times that for a conventional m.o.s. dynamic r.a.m. The design of the basic associative memory cell is described.