Neuromorphic processor and neuromorphic processing method
A neuromorphic processor and a neuromorphic processing method are provided. The neuromorphic processor includes a plurality of neuromorphic elements, a message exchange facility, and a computation unit. The plurality of neuromorphic elements each has a respective state memory entry for storing their state. The message exchange facility enables neuromorphic elements to transmit neural event messages, and to receive transmitted neural event messages. A neural event message includes a message type indicating that the neural event message is an accumulation message or a leakage message. An accumulation message instructs the computation unit to modify a state value of a destination neuromorphic element by addition or subtraction of a value indicated as a message parameter in the accumulation message. A leakage message instructs the computation unit to modify the state value by reduction of the state value with a leakage value that is correlated with the state value.
1 . A neuromorphic processor comprising:
a plurality of neuromorphic elements, each neuromorphic element of the plurality of neuromorphic elements having a respective state memory entry for storing a state of the neuromorphic element;
a message exchange facility to enable the plurality of neuromorphic elements to transmit neural event messages and to receive transmitted neural event messages, wherein a neural event message includes:
an indication of one or more of the plurality of neuromorphic elements selected as a destination, and
a message type; and
a computation unit to update the state of a respective one of the plurality of neuromorphic elements based on a received neural event message received by the respective one of the plurality of neuromorphic elements;
wherein the message type of the received neural event message indicates that the received neural event message is an accumulation message or a leakage message,
wherein the accumulation message instructs the computation unit to modify a state value of the respective one of the plurality of neuromorphic elements by addition or subtraction of a value indicated as a parameter in the accumulation message, and
wherein the leakage message instructs the computation unit to modify the state value by reduction of the state value with a leakage value that is correlated with the state value.
2 . The neuromorphic processor according to claim 1 , wherein the leakage value is equal to a binary shifted difference between the state value and a bias value.
3 . The neuromorphic processor according to claim 2 , wherein a binary shift is specified by a binary shift value specified as a parameter in the leakage message.
4 . The neuromorphic processor according to claim 2 , wherein the bias value is specified as a parameter in the leakage message.
5 . The neuromorphic processor according to claim 2 , wherein the bias value is a predetermined value.
6 . The neuromorphic processor according to claim 1 , wherein the plurality of neuromorphic elements form a neuromorphic cluster of neuromorphic elements, and wherein the plurality of neuromorphic elements share a common memory facility.
7 . The neuromorphic processor according to claim 6 , wherein the neuromorphic cluster is to update the states of the plurality of neuromorphic elements of the neuromorphic cluster on a time-shared basis.
8 . The neuromorphic processor according to claim 1 , wherein each respective neuromorphic element of the plurality of neuromorphic elements comprises a respective computation unit having dedicated access to the respective state memory entry for the respective neuromorphic element.
9 . The neuromorphic processor according to claim 6 ,
wherein the neuromorphic cluster is one of a plurality of neuromorphic clusters, and
wherein each one of the plurality of neuromorphic clusters have access to a message exchange network.
10 . A neuromorphic processing method comprising:
storing a state of each one of a plurality of neuromorphic elements in a respective state memory entry of a plurality of state memory entries;
enabling neuromorphic elements of the plurality of neuromorphic elements to transmit neural event messages, and to receive transmitted neural event messages, a neural event message including an indication of one or more of the plurality of neuromorphic elements selected as a destination, and a message type; and
updating the state of a respective one of the plurality of neuromorphic elements based on a received neural event message received by the respective one of the plurality of neuromorphic elements;
wherein the message type of the received neural event message indicates that the received neural event message is an accumulation message or a leakage message,
wherein the accumulation message causes modifying of a state value of the respective one of the plurality of neuromorphic elements by addition or subtraction of a value indicated as a parameter in the accumulation message, and
wherein the leakage message causes modifying of the state value by reduction of the state value with a leakage value that is correlated with the state value.
11 . The neuromorphic processor according to claim 3 , wherein the bias value is specified as a parameter in the leakage message.
12 . The neuromorphic processor according to claim 3 , wherein the bias value is a predetermined value.
13 . The neuromorphic processor according to claim 1 , wherein the leakage value is correlated with the state value and dependent on one or more parameters contained in the leakage message.
14 . The neuromorphic processing method according to claim 10 , wherein the leakage value is correlated with the state value and dependent on one or more parameters contained in the leakage message.
15 . The neuromorphic processing method according to claim 10 , wherein the leakage value is equal to a binary shifted difference between the state value and a bias value.
16 . The neuromorphic processing method according to claim 15 , wherein a binary shift is specified by a binary shift value specified as a parameter in the leakage message.
17 . The neuromorphic processing method according to claim 15 , wherein the bias value is specified as a parameter in the leakage message.
18 . The neuromorphic processing method according to claim 15 , wherein the bias value is a predetermined value.
19 . The neuromorphic processing method according to claim 10 , wherein the plurality of neuromorphic elements form a neuromorphic cluster of neuromorphic elements, and wherein the plurality of neuromorphic elements share a common memory facility.
20 . The neuromorphic processing method according to claim 19 , comprising updating the states of the plurality of neuromorphic elements of the neuromorphic cluster on a time-shared basis.