A magnetic random access memory chip having magnetic memory elements with different performance characteristics formed on the same chip. The magnetic memory elements can be magnetic random access memory elements. The memory chip can have a first set of magnetic random access chips having a first set of physical and performance characteristics formed in a first area of the sensor and a second set of magnetic random access chips having a second set of performance characteristics formed in a second area of the chip. For example, the first set of magnetic random access memory elements can have performance characteristics that match or exceed those of a non-volatile memory, whereas the second set of magnetic random access memory elements can have performance characteristic that match or exceed those of a static random access memory element.

Embodiments of the present invention provides a system and method of learning and classifying features to identify objects in images using a temporally coded deep spiking neural network, a classifying method by using a reconfigurable spiking neural network device or software comprising configuration logic, a plurality of reconfigurable spiking neurons and a second plurality of synapses. The spiking neural network device or software further comprises a plurality of user-selectable convolution and pooling engines. Each fully connected and convolution engine is capable of learning features, thus producing a plurality of feature map layers corresponding to a plurality of regions respectively, each of the convolution engines being used for obtaining a response of a neuron in the corresponding region. The neurons are modeled as Integrate and Fire neurons with a non-linear time constant, forming individual integrating threshold units with a spike output, eliminating the need for multiplication and addition of floating-point numbers.

Embodiments of the present invention provides a system and method of learning and classifying features to identify objects in images using a temporally coded deep spiking neural network, a classifying method by using a reconfigurable spiking neural network device or software comprising configuration logic, a plurality of reconfigurable spiking neurons and a second plurality of synapses. The spiking neural network device or software further comprises a plurality of user-selectable convolution and pooling engines. Each fully connected and convolution engine is capable of learning features, thus producing a plurality of feature map layers corresponding to a plurality of regions respectively, each of the convolution engines being used for obtaining a response of a neuron in the corresponding region. The neurons are modeled as Integrate and Fire neurons with a non-linear time constant, forming individual integrating threshold units with a spike output, eliminating the need for multiplication and addition of floating-point numbers.

A semiconductor chip is provided. The semiconductor chip includes a SRAM cell, a logic cell, a signal line and a ground line. The SRAM cell includes a storage transmission gate, a read transmission gate and a latch circuit. The latch circuit is serially connected between the storage and read transmission gates, and includes a first inverter, a second inverter and a transmission gate connected to an output of the first inverter, an input of the second inverter and an output of the storage transmission gate. The logic cell disposed aside the SRAM cell is connected with the SRAM cell by first and second active structures. The signal and ground lines extend at opposite sides of the SRAM and logic cells, and are substantially parallel with the first and second active structures. The SRAM and logic cells are disposed between and electrically connected to the signal and ground lines.

A memory device includes a plurality of memory cells located in a first region of the memory device. The memory cells include a first signal line, a first circuit located in the first region of the memory device, and a plurality of logic circuits located in a second region of the memory device. The second region and the first region have different design rules. The first circuit is configured to be selectively enabled and disabled. When the first circuit is enabled, the first signal line is electrically coupled in parallel with a second signal line.

Embodiments of the present invention provides a system and method of learning and classifying features to identify objects in images using a temporally coded deep spiking neural network, a classifying method by using a reconfigurable spiking neural network device or software comprising configuration logic, a plurality of reconfigurable spiking neurons and a second plurality of synapses. The spiking neural network device or software further comprises a plurality of user-selectable convolution and pooling engines. Each fully connected and convolution engine is capable of learning features, thus producing a plurality of feature map layers corresponding to a plurality of regions respectively, each of the convolution engines being used for obtaining a response of a neuron in the corresponding region. The neurons are modeled as Integrate and Fire neurons with a non-linear time constant, forming individual integrating threshold units with a spike output, eliminating the need for multiplication and addition of floating-point numbers.

Embodiments of the present invention provides a system and method of learning and classifying features to identify objects in images using a temporally coded deep spiking neural network, a classifying method by using a reconfigurable spiking neural network device or software comprising configuration logic, a plurality of reconfigurable spiking neurons and a second plurality of synapses. The spiking neural network device or software further comprises a plurality of user-selectable convolution and pooling engines. Each fully connected and convolution engine is capable of learning features, thus producing a plurality of feature map layers corresponding to a plurality of regions respectively, each of the convolution engines being used for obtaining a response of a neuron in the corresponding region. The neurons are modeled as Integrate and Fire neurons with a non-linear time constant, forming individual integrating threshold units with a spike output, eliminating the need for multiplication and addition of floating-point numbers.

A test system is provided that includes a memory test circuit, a memory, an input logic circuit, a bypass circuit, an output logic circuit and a register. The register is operated as a pipeline register of the memory test circuit and the output logic circuit. In a first test mode, the memory test circuit transmits a first test signal to the memory such that the memory outputs a memory output test signal to be stored in the register and further transmitted to the memory test circuit or the output logic circuit to perform test.

An eight-transistor (8T) Static Random-Access Memory (SRAM) cell has four latch transistors, and pairs of n-channel and p-channel pass transistors in parallel to only one pair of bit lines. During read, only the read word line and the n-channel pass transistors are activated, but during a write both the read word line and an extra write word line are activated to turn on all four pass transistors. The cell is powered by VDDM, one threshold above the normal VDD power supply of the read sense and write drivers and interfaces. The bit lines are precharged to VDD but pulled up to VDDM by a latch of cross-coupled p-channel transistors. Any p-channel transistors that connect to the bit lines are driven inactive by VDDM. The read margin is largely decoupled from the write margin by two additional p-channel pass transistors and one extra word line versus a standard 6T cell.

An eight-transistor (8T) Static Random-Access Memory (SRAM) cell has four latch transistors, and pairs of n-channel and p-channel pass transistors in parallel to only one pair of bit lines. During read, only the read word line and the n-channel pass transistors are activated, but during a write both the read word line and an extra write word line are activated to turn on all four pass transistors. The cell is powered by VDDM, one threshold above the normal VDD power supply of the read sense and write drivers and interfaces. The bit lines are precharged to VDD but pulled up to VDDM by a latch of cross-coupled p-channel transistors. Any p-channel transistors that connect to the bit lines are driven inactive by VDDM. The read margin is largely decoupled from the write margin by two additional p-channel pass transistors and one extra word line versus a standard 6T cell.