INTEGRATING MACHINE USING CURRENT MIRROR WITH LOW-VOLTAGE MOSFET, AND NEURAL NETWORK SYNAPSE

    公开(公告)号:JPH10240849A

    公开(公告)日:1998-09-11

    申请号:JP26750197

    申请日:1997-09-30

    Abstract: PROBLEM TO BE SOLVED: To obtain the integrating machine with high performance which is able to operate at low voltage level, and the neural network synapse by using VLSI technology by including a 1st current mirror consisting of MOS transistors to generate a 1st current and a 2nd current mirror which consists of MOS transistors to generate a 2nd current and is coupled with the 1st current mirror in parallel. SOLUTION: An output current I0 fixes a supply voltage V1 and a voltage V3 applied to the gate of an n-channel MOSFET M2 of the 2nd current mirror. When a voltage V2 applied to the gate of an n-channel MOSFET M1 of the 1st current mirror is applied from outside, the final current I0 varies linearly with the current V2 . Namely, the output current I0 does not have a VT term as a secondary term, so this circuit is usable as an integrating machine. Further, elements are constituted in two stages between the supply voltage V1 and a ground voltage GND, so the loss of the voltage is small, so that low-voltage operation becomes possible.

    HIGH-SPEED PACKET SWITCH CONTROLLER USING NEURAL NETWORK CHIP AND EXCHANGE USING SAME

    公开(公告)号:JPH11284662A

    公开(公告)日:1999-10-15

    申请号:JP29515098

    申请日:1998-10-16

    Abstract: PROBLEM TO BE SOLVED: To provide the high-speed packet switch controller which is very suitable to a large-capacity packet switch controller and can maximize system performance through optimum switching by making use of the neural network chip. SOLUTION: A connection ring array 42 to which addresses generated by decoding a weighted value row address and a weighted value column address through a weighted value row address decoder 40 and a weighted value column address decoder 41 are inputted coupled an input signal to a neural network 43 according to the said address and also varies an output value according to an externally inputted weighted value. The neural network 43 outputs a crossbar switch control signal, generated by amplifying the signal outputted from the connection ring array 42, to the outside through an input/output bus 44.

Patent Agency Ranking