Charge domain synapse cell

Claim 2

A charge domain synapse cell comprising: an input node; first and second summing lines having associated first and second potentials, respectively; a capacitor coupled between an intermediate node and said input node; a first device means having a programmed voltage threshold for transferring a packet of charge from said second summing line to said intermediate node in response to a high-to-low voltage transition at said node; a second device means for transferring said charge packet from said intermediate node to said first summing line in response to the next low-to-high voltage transition following said high-to-low transition, at said input node.

Claim 3

The synapse cell of claim 1 wherein the magnitude of said charge packet transferred is a function of the capacitance of said capacitor, the magnitude of said transitions and said programmed voltage threshold.

Claim 4

The synapse cell of claim 2 wherein at least said second summing line is initially charged to a reference potential.

Claim 5

The synapse cell of claim 3 wherein said first device means comprises a floating gate device programmed to hold a charge to establish said voltage threshold, said charge representing a stored weight.

Claim 6

The synapse cell of claim 4 wherein said second device means comprises a field-effect device.

Claim 7

The synapse cell of claim 5 further comprising a means for programming said floating gate device.

Claim 8

The synapse cell of claim 6 wherein said first and said summing lines are differentially coupled to the input of a neural summing device.

Claim 9

The synapse cell of claim 4 wherein application of a fixed number of voltage pulses to said input node transfers a total charge from said second summing line to said first summing line which is proportional to the product of said fixed number of pulses, said weight and said capacitance.

Claim 10

A neural network comprising a plurality of input wordlines and a corresponding plurality of column summing lines, pairs of said column summing lines being differentially coupled to an associated neural summing device, said network further comprising an array of synapse cells providing weighted connections between said input wordlines and said column summing lines, each of said synapse cells comprising: a programmable floating gate device having a floating gate, a control gate and said first and second spaced-apart doped regions formed in a semiconductor substrate, said first doped region being coupled to a first column summing line charged to a reference potential, with a second doped region being coupled to a charging node, said floating gate storing a charge which establishes the threshold of said floating gate device, said charge corresponding to a weight; a semiconductor device for coupling said charging node to a second column summing line, said first and second column summing lines comprising one of said pairs; a capacitance coupled between said charging node and an input wordline, said capacitance coupling voltage transitions on said input wordline to said charging node; each cell being operational such that whenever a high-to-low voltage transition occurs on said wordline which causes the voltage at said charging node to drop below said reference potential minus said threshold, a packet of charge is transferred from said first column summing line to said charging node said charge packet, or a fractional portion thereof, being transferred to said second column summing line in response to subsequent low-to-high transition on said input wordline.

Claim 11

A neural network comprising a plurality of input wordlines and a corresponding plurality of column summing lines, pairs of said column summing lines being differentially coupled to an associated neural summing device, said network further comprising an array of synapse cells providing weighted connections between said input wordlines and said column summing lines, each of said synapse cells comprising: a programmable floating gate device having a floating gate, a control gate and said first and second spaced-apart doped regions formed in a semiconductor substrate, said first doped region being coupled to a first column summing line charged to a reference potential, with a second doped region being coupled to a charging node, said floating gate storing a charge which establishes the threshold of said floating gate device, said charge corresponding to a weight; a semiconductor device for coupling said charging node to a second column summing line, said first and second column summing lines comprising one of said pairs; a capacitance coupled between said charging node and an input wordline, said capacitance coupling voltage transitions on said input wordline to said charging node; each cell being operational such that whenever a high-to-low voltage transition occurs on said wordline which causes the voltage at said charging node to drop below said reference potential minus said threshold, a packet of charge is transferred from said first column summing line to said charging node said charge packet, or a fractional portion thereof, being transferred to said second column summing line in response to subsequent low-to-high transition on said input wordline.

Claim 12

The network of claim 9 wherein the amount of charge transferred from said first to said second column summing lines is proportional to the product of said weight, the amplitude of said transitions and said capacitance.

Claim 13

The network of claim 10 further comprising a means for programming said floating gate device.

Claim 14

The network of claim 11 wherein said semiconductor device comprises a first field-effect device configured as a diode so that current flow is directional from said charging node to said second column summing line.

Claim 15

The network of claim 12 wherein the total charge transferred from said first to said second column summing lines during a certain time period is proportional to the product of the number of said transitions occurring within said time period, said amplitude of said transitions, said weight and said capacitance.

Claim 16

The network of claim 13 wherein said programming means comprises: a second field-effect device coupled in series between said charging node and said input wordline, the gate of said second field-effect device being coupled to a first programming line such that when said first programming line is raised to a high positive potential, said second field-effect device connects said input wordline directly to the drain of said floating gate device; a third field-effect device coupled in series between said first doped region of said floating gate device and said first column summing line, the gate of said third field-effect device being coupled to a second programming line such that when said second programming line is grounded, the control gate of said floating gate device is disconnected from said second region, yet said control gate remains coupled to said first column summing line.

Claim 17

The network of claim 14 wherein said neural summing device implements a Sigmoid function.

Claim 18

A semiconductor synapse cell for transferring charge from a first to a second summing line in response to an input voltage pulse comprising: a capacitor means for coupling the voltage transitions associated with said pulse to an internal node within said cell; first semiconductor device means for transferring a packet of charge from said first summing line to said internal node in response to a first transition of said pulse; and second semiconductor device means for transferring said charge packet, or a fraction thereof, to said summing line in response to a second transition of said pulse.

Claim 19

A semiconductor synapse cell for transferring charge from a first to a second summing line in response to an input voltage pulse comprising: a capacitor means for coupling the voltage transitions associated with said pulse to an internal node within said cell; first semiconductor device means for transferring a packet of charge from said first summing line to said internal node in response to a first transition of said pulse; and second semiconductor device means for transferring said charge packet, or a fraction thereof, to said summing line in response to a second transition of said pulse.

Claim 20

The synapse cell of claim 16 wherein said first semiconductor device means comprises a first field-effect device.

Claim 21

The cell of claim 17 wherein said first field-effect device comprises a floating gate device.

Claim 22

The synapse cell of claim 18 wherein said first semiconductor device means further comprises a second field-effect device coupled in series with said first field-effect device, said second field-effect device being configured as a diode.

Claim 23

The synapse cell of claim 19 wherein said second semiconductor device means comprises a third field-effect device configured as a diode.

Claim 24

The synapse cell of claim 17 further comprising means for programming said first field-effect device.

Claim 25

The synapse cell of claim 21 wherein said programming means comprises a fully-decoded EEPROM cell.

Claim 26

The synapse cell of claim 18 further comprising means for programming said floating gate device.