KR101220481B1 - Networks on chip system and method of routing thereof - Google Patents

Abstract

The network-on-chip system includes a plurality of layer portions each including at least one IP module, at least one local router connected to each IP module, and a parallel router connected to the local router, wherein the parallel routers are connected in parallel to a common bus and have different layers. Sends and receives data packets with a negative parallel router in a broadcast manner.

Description

Network on chip system and its routing method {NETWORKS ON CHIP SYSTEM AND METHOD OF ROUTING THEREOF}

The present invention relates to a network-on-chip system and a data routing method thereof.

As the integration of circuits increases with the development of semiconductor technology, billions of transistors can be integrated in one chip. In addition, as the demands of consumers vary, system-on-chips (SoCs) with dozens or hundreds of IPs are used to satisfy these requirements.

When such a large number of IPs are integrated in a single chip, the shared bus system, which is a representative structure for communication within the chip, increases the probability of bottlenecks as the number of IPs increases. To solve this problem, a structure called network-on-chip (NoC) has been studied.

Meanwhile, two-dimensional integrated circuits in which devices are disposed in a plan have disadvantages such as increased delay and increased power consumption as the number of internal devices increases. The three-dimensional integrated circuits studied to solve this problem improve the system performance by reducing the short communication distance and power consumption. On the basis of these advantages, research on 3D NoC combining 3D integrated circuit technology and NoC technology is being actively conducted.

However, the conventional three-dimensional NoC structure is a symmetric NoC (Symmetric NoC), the shared bus system is a structure that passes through all the layers (Layer). Accordingly, there is a problem in that a plurality of hops are consumed as data passes through a plurality of layers in inter-layer communication.

In order to solve the above problems of the prior art, an embodiment of the present invention is to provide a three-dimensional network on a chip system and a routing method that can reduce the data transmission delay.

Network on a chip system according to an aspect of the present invention for achieving the above technical problem, at least one IP (Intellecture Property) module, at least one local router connected to each of the IP module, and parallel to the local router A plurality of layer units each comprising a router, wherein the parallel router is connected in parallel to a common bus and transmits and receives data packets in parallel with a parallel router of another layer unit.

In addition, a parallel routing device each configured in a plurality of layer units on a network-on-chip system according to another aspect of the present invention and connected in parallel with a common bus may include another layer unit. A data input unit for receiving an input data packet generated from an IP (Intellecture Property) module through the common bus and confirming a destination address of the input data packet; If the destination address of the input data packet matches the address of the parallel routing device, send the input data packet to a destination local router according to the destination address; and if the destination address does not match the address of the parallel routing device, the input A path controller for drop processing a data packet; And temporarily storing the output data packet generated from the IP module of the layer unit in which the parallel routing device is configured, and when the right to use the common bus is obtained, broadcast the output data packet to another parallel routing device through the common bus. cast) includes a data output unit for transmitting, wherein the layer unit includes at least one IP module and at least one local router connected to the IP module.

In addition, a parallel routing method for a plurality of layer units connected in parallel to a common bus on a network-on-chip system according to another aspect of the present invention may include any layer. Receiving an input data packet generated from an IP (Intellecture Property) module included in another layer unit through the common bus; And transmitting the input data packet to a destination local router included in the arbitrary layer part when the destination address of the input data packet matches the address of the arbitrary layer part, wherein the layer part includes at least one It includes an IP module and at least one local router connected to the IP module.

In addition, a parallel routing method for a plurality of layer units connected in parallel to a common bus on a network-on-chip system according to another aspect of the present invention may include any layer. Temporarily storing an output data packet generated from an IP (Intellecture Property) module included in the unit; Determining acquisition of a usage right of the common bus for the arbitrary layer unit; And if the right to use the common bus is obtained, sequentially outputting the output data packet to the shared bus and transmitting the broadcast data to another layer unit, wherein the layer unit includes: at least one IP module; At least one local router connected to the IP module.

According to any one of the above-described means for solving the problem of the present invention, by providing a three-dimensional network on a chip system in which a plurality of layers are connected in parallel to a common bus, the data transfer delay time by reducing the hop during data transfer between layers There is an effect that can be reduced.

1 is a view showing the structure of a NoC system according to an embodiment of the present invention.
2 is a diagram illustrating an IP topology structure according to an embodiment of the present invention.
3 is a diagram illustrating an IP topology structure according to another embodiment of the present invention.
4 is a block diagram showing the structure of a parallel router according to an embodiment of the present invention.
5 is a diagram illustrating a data packet structure in a NoC system according to an embodiment of the present invention.
6 is a flowchart illustrating a parallel routing method for input data according to an embodiment of the present invention.
7 is a flowchart illustrating a parallel routing method for output data according to an embodiment of the present invention.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings, which will be readily apparent to those skilled in the art. The present invention may, however, be embodied in many different forms and should not be construed as limited to the embodiments set forth herein. In the drawings, parts irrelevant to the description are omitted in order to clearly describe the present invention, and like reference numerals designate like parts throughout the specification.

Throughout the specification, when a part is referred to as being "connected" to another part, it includes not only "directly connected" but also "electrically connected" with another part in between . Also, when an element is referred to as "comprising ", it means that it can include other elements as well, without departing from the other elements unless specifically stated otherwise.

1 is a view showing the structure of a NoC system according to an embodiment of the present invention.

2 is a diagram illustrating an IP topology structure in a NoC system according to an embodiment of the present invention, and FIG. 3 is a diagram illustrating an IP topology structure in a NoC system according to another embodiment of the present invention.

As shown in FIG. 1, in a NoC system in a multi-core processor environment according to an exemplary embodiment of the present invention, a plurality of layers (Layer 1, 2, 3... N) are configured in a parallel structure.

Specifically, each layer (Layer 1, 2, 3 ...... n) part includes a plurality of IP (Intellcture Property) module (hereinafter referred to as "IP"), each IP is a local router ) In addition, a plurality of local routers configured for each layer unit are connected to one parallel router, and the parallel routers of each layer unit are connected in parallel to one common bus 40.

In this case, the layer unit may be connected to a plurality of IP and local routers in the same structure as in FIGS. 2 and 3.

First, as shown in FIG. 2, the IPs 10 in the NoC system according to the exemplary embodiment of the present invention are connected one-to-one with the local router (LR) 20, respectively, and the plurality of IPs 10 The LRs 20 are connected in a mesh topology structure.

For example, FIG. 2 shows that the nineth IP 10 and the nine LRs 20 are included in the n-th layer part among n layer parts, and the nine LRs 20 are all higher. The connection with the PR 30, which is a router, is shown. As shown in FIG. 1, the PR 30 is connected to one common bus 40 in parallel to transmit and receive data to and from the PR 30 of other layer units.

In addition, as shown in FIG. 3, in the NoC system according to another embodiment of the present invention, a plurality of IPs 10 included in each layer unit are grouped into a plurality of groups consisting of a predetermined number of IPs 10. In addition, the IP 10 included in each group may be configured in a star topology structure connected to one LR 20. Then, LRs 20 of each group are connected to one PR 30.

For example, FIG. 3 shows that 16 IPs 10 are included in an nth layer part among n layer parts, and 4 IPs 10 are grouped into one group. . At this time, the IP 10 of each group is connected to one LR 20, the four LR 20 is connected in a ring topology structure to route data transmitted and received between the IP (10). Four LRs 20 are all connected to one PR 30. As shown in FIG. 1, the PR 30 is connected to one common bus 40 in parallel to transmit and receive data to and from the PR 30 of other layer units.

 Meanwhile, in the NoC system according to the embodiment of the present invention, the topology structure of the IP 10 and the LR 20 configured in each layer part may be applied in addition to the IP topology structures shown in FIGS. 2 and 3. have.

And, in the NoC system according to an embodiment of the present invention, the number of IPs and LRs included in one layer unit may be configured differently or identically for each layer unit.

In addition, the IP and the LRs each include a plurality of input / output ports, and can input / output data through the input / output ports. At this time, the LR 20 serves as a routing switch for transmitting data generated from the connected IP 10 to the destination IP.

As such, the NoC system according to an embodiment of the present invention performs a three-dimensional routing process. That is, horizontal routing processing is performed between the LRs 20 configured in each layer unit, and vertical (ie, parallel) routing processing is performed between the parallel routers 30 respectively included in the plurality of layer units.

Meanwhile, in FIGS. 3 and 4, in the NoC system according to an embodiment of the present invention, the PR 30 is connected one-to-one with the LRs 20 of the corresponding layer part. However, the PR 30 is based on a horizontal communication method. It is also possible to connect only to the LR 20 of the last stage (ie the last stage before the common bus). As such, by only connecting the LR 20 of the last stage, the number of input / output ports of the PR 30 can be reduced. At this time, the PR 30 and the LR 20 of each layer unit may route data in an X-Y routing manner.

Meanwhile, FIGS. 1 to 3 show that a plurality of LRs 20 and one PR 30 are configured per layer part in the NoC system according to an embodiment of the present invention. However, in the NoC system according to another embodiment of the present invention, each LR 20 may be configured as one of the PR 30 for each layer unit. That is, a plurality of "system routers" including a module that functions as a PR 30 for each LR 20 connected to each IP 10 per one layer part are included. In this case, the input and output ports of the module functioning as the PR 30 in the system router are connected in parallel to a common bus with other layer units, respectively.

Hereinafter, a parallel routing apparatus and a parallel routing method on a NoC system according to an embodiment of the present invention will be described in detail with reference to FIGS. 4 to 7.

4 is a block diagram showing the structure of a parallel router according to an embodiment of the present invention.

5 is a diagram illustrating a data packet structure in a NoC system according to an embodiment of the present invention.

First, as shown in FIG. 4, the parallel router 30 according to an embodiment of the present invention includes a data input unit 301, an input data buffer 302, a path controller 303, an output data buffer 304, And a data output unit 305.

The data input unit 301 sequentially receives input data generated from the IP of another layer unit through the common bus 40. The data input unit 301 checks the destination address of the input data and transfers it to the path controller 303.

For reference, in the NoC system 10 according to an exemplary embodiment of the present invention, data transmitted and received between each layer unit may have a structure including a plurality of fields in a packet form.

For example, as shown in (a) of FIG. 5, a data packet according to an embodiment of the present invention has a 'source address' field in which source address information of an IP where a corresponding data packet is generated is stored. A 'destination address' field (P50) in which address information of the destination parallel router and destination local router is stored, a 'type' field in which start and end information indicating the order of the corresponding data packet is stored, and a 'payment' in which actual data is stored. It has a payload field structure.

In this case, as shown in FIG. 5B, the 'destination address' field P50 may include a 'parallel address' field P51 and a 'horizontal address' field P52. In the parallel address field P51, a parallel router address for selecting a specific layer part as a destination among a plurality of layer parts connected in parallel is stored, and the horizontal address field P52 is horizontal in one layer part. The local router address for selecting a specific local router as a destination among a plurality of local routers connected to the server is stored.

5, the input data buffer 302 sequentially stores the input data destined for the layer unit among the input data received through the data input unit 301 under the control of the path controller 303.

The input data buffer 302 sequentially transmits the temporarily stored input data to the corresponding destination local router under the control of the path controller 303.

The path controller 303 checks the destination address of the input data packet, and if the destination parallel router address matches the address of the parallel router, transmits the input data packet to the destination local router, and the destination address matches the address of the parallel router. Otherwise, the input data packet is dropped.

In this case, the path controller 303 stores the input data packet whose destination address is the corresponding parallel router among the input data packets in the input data buffer 302 and sequentially outputs the input data packet to the corresponding destination local router side.

The output data buffer 304 temporarily stores output data generated from any one of IPs included in the layer unit under the control of the data output unit 305. The output data buffer 304 stores a dummy data packet in advance, and the dummy data packet includes end information indicating the end of the use right of the common bus 40. For reference, the parallel routers receiving the dummy data packet determine that the parallel router transmitting the dummy data packet has terminated the right to use the common bus 40.

The data output unit 305 temporarily stores the output data packet generated from the IP of the layer unit in which the corresponding parallel router is configured in the output data buffer. When the data output unit 305 acquires the right to use the common bus 40, the data output unit 305 outputs the temporarily stored output data packet through the common bus 40. That is, the output data packet output through the parallel router is broadcasted to the parallel router of another layer part.

At this time, the data output unit 305 acquires a preset dummy data packet for notification of termination of usage right from the output data buffer 304 when there is no temporarily stored output data packet when the usage right of the common bus is acquired. 40).

Meanwhile, in the NoC system according to an exemplary embodiment of the present invention, a token bus in which a plurality of parallel routers 30 sequentially acquires the right to use the common bus 40 is based on a predetermined common bus use right acquisition order. bus method is applied.

In this case, the data output unit 305 is received from the parallel routers in the immediately preceding order of the corresponding parallel routers in the preset order, and determines whether to obtain the right to use the common bus by determining whether the received input data packet includes termination information. do.

For reference, as shown in FIG. 5, the 'type' field of the input data packet stores 'end information', which means that the input data packet is the last packet transmitted from the parallel router of the layer unit.

Hereinafter, a parallel routing method in a NoC system according to an embodiment of the present invention will be described in detail with reference to FIGS. 6 and 7.

6 is a flowchart illustrating a parallel routing method for input data according to an embodiment of the present invention.

First, when data input is generated through the common bus (S610), the destination address included in the input data is checked (S620).

In this case, the input data is a data packet generated from an IP included in another layer unit and broadcast through a common bus.

Thereafter, it is determined whether the parallel address (that is, the destination parallel router address) included in the checked destination address matches the corresponding layer part (S630).

For reference, the matching may be determined by comparing the parallel address with the address of the parallel router.

When the parallel address matches the layer unit as a result of the determination in step S630, the input data is transmitted to the destination local router side based on the horizontal address included in the destination address (S640).

In this case, the input data may be horizontally routed (for example, X-Y routing) to a destination local router through a local router before the parallel router of the layer unit, or may be directly routed to a destination local router connected one-to-one to the parallel router.

On the other hand, if the parallel address does not match with the corresponding layer part as a result of the determination in S630, the input data is dropped (S650).

7 is a flowchart illustrating a parallel routing method for output data according to an embodiment of the present invention.

First, when a common bus use right for the layer unit is generated (S710), it is determined whether there is pre-stored or generated output data (S720).

In this case, the output data is data generated from any one of IPs of the corresponding layer unit. If a data packet including transmission termination information is received from a layer unit of a previous sequence among the preset sequences, it is determined that a common bus usage right has been generated for the layer unit.

If it is determined in step S720 that the output data exists, the corresponding output data is output to the common bus 40 and broadcasted to another parallel router (S730).

In this case, at least one output data may be sequentially output, and each output data includes destination address information and type information.

On the other hand, if it is determined in step S720 that the output data does not exist, the preset dummy data is output to the common bus and broadcasted to another parallel router (S740).

In this case, the dummy data includes termination information on the common bus usage right acquired by the parallel router of the layer unit.

The foregoing description of the present invention is intended for illustration, and it will be understood by those skilled in the art that the present invention may be easily modified in other specific forms without changing the technical spirit or essential features of the present invention. will be. It is therefore to be understood that the above-described embodiments are illustrative in all aspects and not restrictive. For example, each component described as a single entity may be distributed and implemented, and components described as being distributed may also be implemented in a combined form.

The scope of the present invention is shown by the following claims rather than the above description, and all changes or modifications derived from the meaning and scope of the claims and their equivalents should be construed as being included in the scope of the present invention. do.

10: IP 20: local router
30: parallel router 40: common bus

Claims (17)

In a network-on-chip system,
At least one IP (Intellecture Property) module,
At least one local router connected to each of the IP modules, and
Including a plurality of layer (layer) portion each including a parallel router connected to the local router,
The parallel router,
It is connected in parallel to a common bus and sends and receives data packets in parallel with other layer parallel routers.
If the destination address of the input data packet received through the common bus is the address of the parallel router, the input data packet is transmitted to the destination local router connected to the parallel router, and if the destination address of the input packet is the address of another parallel router A network on chip system for drop processing the input data packet. The method of claim 1,
The parallel router,
The data packet is transmitted through a common bus in a token bus manner,
The token bus method is a network on a chip system after the parallel router in a predetermined order to terminate the right to use the common bus, the next parallel router to obtain the right to use the common bus. 3. The method according to claim 1 or 2,
The data packet is,
And at least one of a source address field, a destination address field, a type information field, and a payload field,
The destination address field stores a destination parallel router address and a destination local router address;
And the type information field stores start and end information of the data packet. The method of claim 3, wherein
The parallel router,
And receiving a data packet including the termination information from a parallel router in a previous order of a preset sequence, and determining that the right to use the common bus has been acquired. The method of claim 1,
The parallel router,
After acquiring the usage right of the common bus, output a valid data packet to the common bus,
And outputting a preset dummy data packet for end of use right notification to the common bus when the valid data packet does not exist. delete The method of claim 1,
The IP module and local routers are connected in a mesh topology structure or a star topology structure. In a parallel routing device each configured in a plurality of layer units on a network-on-chip system and connected in parallel with a common bus,
A data input unit for receiving an input data packet generated from an IP (Intellecture Property) module of another layer unit through the common bus and confirming a destination address of the input data packet;
If the destination address of the input data packet matches the address of the parallel routing device, send the input data packet to a destination local router according to the destination address; and if the destination address does not match the address of the parallel routing device, the input A path controller for drop processing a data packet; And
Temporarily storing the output data packet generated from the IP module of the layer unit configured to the parallel routing device, if the right of use of the common bus is obtained, broadcast the output data packet to another parallel routing device through the common bus (broad cast) ) Includes a data output unit for transmitting,
The layer unit,
And at least one IP module and at least one local router coupled to the IP module. The method of claim 8,
The data output unit,
And, when the output data packet does not exist, outputting a preset dummy data packet for end of use right notification to the common bus when the use right of the common bus is acquired. The method of claim 8,
The input and output data packet,
And at least one of a source address field, a destination address field, a type information field, and a payload field,
The destination address field stores a destination parallel routing device address and a destination local router address;
And the type information field stores start and end information of the data packet. 11. The method of claim 10,
The data output unit,
And judging whether an input data packet received through the data input unit is received from a parallel routing device in a previous order of a preset order and including the termination information to determine acquisition of a right to use the common bus. The method of claim 8,
Wherein the IP modules and local routers are connected in a mesh or star topology structure. In the parallel routing method for a plurality of layer units connected in parallel to each common bus on a network-on-chip system,
Receiving, by an arbitrary layer unit, an input data packet generated from an IP (Intellecture Property) module included in another layer unit through the common bus; And
If the destination address of the input data packet matches the address of the arbitrary layer portion, the input data packet is transmitted to the destination local router included in the arbitrary layer portion, and the destination address of the input data packet and the arbitrary layer. If the negative address does not match, dropping the input data packet;
The layer unit,
And at least one IP module and at least one local router coupled to the IP module. delete In the parallel routing method for a plurality of layer units connected in parallel to each common bus on a network-on-chip system,
Temporarily storing an output data packet generated from an IP (Intellecture Property) module included in an arbitrary layer unit;
Determining acquisition of a usage right of the common bus for the arbitrary layer unit; And
When acquiring the right to use the common bus, and outputting the output data packet to the common bus in sequence to send a broadcast (broad cast) to another layer unit,
The layer unit,
And at least one IP module and at least one local router coupled to the IP module. The method of claim 15,
In determining the acquisition of the use rights of the common bus,
And determining that the right to use the common bus has been acquired when a data packet including termination information is received from the layer unit in the immediately preceding order of the predetermined layer unit in a predetermined order. The method of claim 15,
Prior to the step of temporarily storing the output data packet,
Determining acquisition of a usage right of the common bus for the arbitrary layer unit; And
And if the right to use the common bus is acquired, broadcasting a preset dummy data packet for end of use right notification to another layer unit through the common bus.

Images