Miscellaneous active electrical nonlinear devices – circuits – and – Signal converting – shaping – or generating – Converting input voltage to output current or vice versa
Reexamination Certificate
2003-05-29
2004-10-26
Lam, Tuan T. (Department: 2816)
Miscellaneous active electrical nonlinear devices, circuits, and
Signal converting, shaping, or generating
Converting input voltage to output current or vice versa
C706S033000
Reexamination Certificate
active
06809558
ABSTRACT:
FIELD OF THE INVENTION
This invention relates to an output neuron circuit and, more particularly, to a novel push-pull output neuron circuit.
BACKGROUND OF THE INVENTION
One component of an artificial neural network is its neurons, whose performance and complexity greatly affect the network. In many cases, the transfer function of the neuron is sigmoidal, and a differential amplifier has generally been used to generate such a transfer function.
In some cases, both the sigmoidal transfer function and its derivative are required, and neurons that meet such a requirement have been developed. Most of these neurons have voltage inputs and current outputs. However, a neuron with current inputs and voltage outputs would be more convenient in some applications, for example, those that employ current output synapses and voltage output neurons, because such a neuron may be capable of summing multiple input currents by connecting the input currents together in parallel at an input terminal of the neuron. U.S. Pat. No. 6,429,699 of Bingxue Shi and Chun Lu, inventors in the present application, discloses a neuron that receives a current signal and outputs a voltage.
SUMMARY OF THE INVENTION
In accordance with the present invention, there is provided a neuron circuit for generating a transfer function of a current input signal that includes a converter to convert the current input signal into a voltage signal, the converter including a current source connectable to a first bias voltage, wherein a current generated by the current source is adjustable by the first bias voltage, a resistive load connectable to at least a second bias voltage and coupled to the current source, wherein a resistance of the resistive load is adjustable by the second bias voltage, wherein the current input signal is coupled to the resistive load and is in parallel with the current generated by the current source, and the converter outputs the voltage signal at a node between the current source and the resistive load, and at least one output circuit coupled to the converter to output the transfer function, the output circuit including a differential amplifier and a current mirror as an active load for the differential amplifier.
Also in accordance with the present invention, neuron circuit for generating a transfer function of a current input signal that includes a converter to convert the current input signal Into a voltage signal, the converter including a first current source connectable to a first bias voltage, wherein a first current generated by the first current source is adjustable by the first bias voltage, a resistive load coupled to the first current source, comprising a PMOS transistor connectable to a second bias voltage and an NMOS transistor connectable to a third bias voltage, a resistance of the resistive load being adjustable by the second bias voltage and the third bias voltage, wherein the current input signal is coupled to the resistive load and is in parallel with the first current, and an output circuit coupled to the converter to output the transfer function, the output circuit including a second current source connectable to a fourth bias voltage, wherein a second current generated by the second current source is adjustable by the fourth bias voltage, a differential amplifier biased by the second current source, and a current mirror as an active load of the differential amplifier.
Still in accordance with the present invention, there is provided a neuron circuit for generating a transfer function of a current input signal and a derivative of the transfer function that includes a converter to convert the current input signal into a voltage signal that includes a first current source connectable to a first bias voltage, wherein a first current generated by the first current source is adjustable by the first bias voltage, a resistive load comprising a PMOS transistor connectable to a second bias voltage and an NMOS transistor connectable to a third bias voltage, a resistance of the resistive load being adjustable by the second bias voltage and the third bias voltage, wherein the current input signal is coupled to the resistive load and is in parallel with the first current, a first output circuit coupled to the converter that includes a second current source connectable to a fourth bias voltage, wherein a second current generated by the second current source is adjustable by the fourth bias voltage, a first differential amplifier biased by the second current source, comprising a first pair of MOS transistors, and a first current mirror as an active load of the first differential amplifier, and a second output circuit coupled to the converter that includes a third current source connectable to a fifth bias voltage, wherein a third current generated by the third current source is adjustable by the fifth bias voltage, a second differential amplifier biased by the third current source, comprising a second pair of MOS transistors, and a second current mirror as an active load of the second differential amplifier, wherein an output of a first one of the first and second output circuits is the transfer function, and a difference between an output of the first output circuit and an output of the second output circuit is the derivative of the transfer function.
Further in accordance with the present invention, there is provided a method for generating a sigmoid transfer function for a neuron circuit that receives a current input signal that includes providing a means to convert the current input signal into a voltage signal, and providing a means for generating the sigmoid transfer function using a differential amplifier and a current mirror as an active load.
Still further in accordance with the present invention, there is provided a method for generating a sigmoid transfer function and a derivative of the sigmoid transfer function for a neuron circuit that receives a current input signal that includes converting the current input signal into a voltage signal, generating a first sigmoid transfer function using a first differential amplifier and a first current mirror as an active load for the first differential amplifier, generating a second sigmoid transfer function using a second differential amplifier and a second current mirror as an active load for the second differential amplifier, and taking one of the first and second sigmoid transfer functions as the sigmoid transfer function for the neuron circuit and generating the derivative of the sigmoid transfer function by comparing the first and second sigmoid transfer functions.
Additional features and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. The features and advantages of the invention will be realized and attained by means of the elements and combinations particularly pointed out in the appended claims.
It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory and are intended to provide further explanation of the invention as claimed.
REFERENCES:
patent: 6429699 (2002-08-01), Shi et al.
patent: 6483383 (2002-11-01), Wu
patent: 6664818 (2003-12-01), Shi et al.
patent: 6687686 (2004-02-01), Nervegna et al.
Chun Lu and bingxue Shi, “Circuit design of an adjustable neuron activation function and its derivative”, Electronics Letters, vol. 36, No. 6, Mar. 16, 2000, pp. 553-555.
Chen Lu
Lu Chun
Shi Bingxue
Finnegan Henderson Farabow Garrett & Dunner L.L.P.
Lam Tuan T.
Winbond Electronics Corporation
LandOfFree
Push-pull output neuron circuit does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Push-pull output neuron circuit, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Push-pull output neuron circuit will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3281705