Data transmission method based on UCIe, artificial intelligence chip, medium and electronic device

Abstract

The invention provides a UCIe-based data transmission method, an artificial intelligence chip, a medium and an electronic device. The UCIe-based data transmission method comprises the following steps of: communicating with a module partner end of a second core grain through a module end of a first core grain; when the module partner end receives the retraining trigger signal, resetting a circuit of the module partner end, and sending a retraining request signal to the module partner end through the module end; when the module partner end receives the retraining request signal, replying a retraining response signal to the module end through the module partner end; and when the module end receives the retraining response signal, enabling the module end to enter a retraining state. According to the UCIe-based data transmission method, the artificial intelligence chip, the medium and the electronic device, good data transmission quality can be provided.

Description

Technical Field

[0001] The present invention relates to the field of chip technology, and in particular to a UCIe-based data transmission method, artificial intelligence chip, medium, and electronic device. Background Art

[0002] Generally speaking, auxiliary channels between chiplets (such as clock signal lines) may generate glitches during signal transmission due to power supply noise and crosstalk between lines, leading to errors in auxiliary channel information parsing. For example, during real-time calibration or retraining, if auxiliary channel message parsing errors occur, the message exchange between the module and its module partner will time out, causing chiplet reporting interruptions. Summary of the Invention

[0003] The present invention is directed to a UCIe-based data transmission method, artificial intelligence chip, medium, and electronic device, which can achieve good data transmission quality.

[0004] According to an embodiment of the present invention, the UCIe-based data transmission method of the present invention includes the following steps: communicating with the module partner end of the second chip through the module end of the first chip; when the module partner end receives a retraining trigger signal, resetting the circuit of the module partner end, and sending a retraining request signal to the module partner end through the module end; when the module partner end receives the retraining request signal, replying a retraining response signal to the module end through the module partner end; and when the module end receives the retraining response signal, causing the module end to enter a retraining state.

[0005] In the UCIe-based data transmission method according to an embodiment of the present invention, the following steps are also included: when the module end enters the active state, the counter starts counting; and when the counter count reaches a default value, in response to an auxiliary channel message parsing error, an interrupt is triggered and the counter is restarted.

[0006] In the UCIe-based data transmission method according to an embodiment of the present invention, the following steps are also included: when the counter count reaches a default value, a recalibration tracking mode request signal is sent to the module partner end through the module end; and when the module partner end parses the recalibration tracking mode request signal incorrectly, an interrupt is triggered through the module partner end, and the counter is restarted.

[0007] In the UCIe-based data transmission method according to an embodiment of the present invention, the following steps are also included: when the module partner end successfully parses the recalibration tracking mode request signal, the module partner end sends a recalibration tracking mode response signal to the module end; and when the module end parses the recalibration tracking mode response signal incorrectly, an interrupt is triggered through the module end, and the counter is reset.

[0008] The UCIe-based data transmission method according to an embodiment of the present invention further includes the following steps: when the module partner successfully parses the recalibration tracking mode request signal, the module partner starts sending the tracking mode signal and the clock mode signal.

[0009] In the UCIe-based data transmission method according to an embodiment of the present invention, the following steps are also included: when the module side successfully parses the recalibration tracking mode response signal, the module side executes the tracking recalibration algorithm; and when the tracking recalibration algorithm is completed, the module side sends a recalibration tracking mode completion signal to the module partner side.

[0010] The UCIe-based data transmission method according to the embodiment of the present invention further includes the following steps: when the module partner end completes signal parsing error for the recalibration tracking mode, triggering an interrupt through the module partner end and re-counting the counter.

[0011] In the UCIe-based data transmission method according to an embodiment of the present invention, the following steps are also included: when the module partner end successfully parses the recalibration tracking mode completion signal, the module partner end sends a recalibration tracking mode completion confirmation signal to the module end; and when the module end parses the recalibration tracking mode completion confirmation signal incorrectly, an interrupt is triggered through the module end, and the counter is restarted.

[0012] In the UCIe-based data transmission method according to an embodiment of the present invention, the following steps are also included: when the module end enters the active state, the module partner end directly starts to send the tracking mode signal and the clock mode signal to the module end, and the counter starts counting; when the counter count reaches a default value, the tracking recalibration algorithm is executed by the module end; and when the tracking recalibration algorithm is executed, the counter is restarted.

[0013] The UCIe-based data transmission method according to the embodiment of the present invention further includes the following steps: when a glitch occurs in the clock channel between the module end and the module partner end, an auxiliary channel message parsing error occurs.

[0014] According to an embodiment of the present invention, an artificial intelligence chip of the present invention includes a first core particle and a second core particle. The first core particle has a module end. The second core particle is coupled to the first core particle and has a module partner end. When the module partner end receives a retraining trigger signal, the circuit of the module partner end is reset, and the module end sends a retraining request signal to the module partner end. When the module partner end receives the retraining request signal, the module partner end replies with a retraining response signal to the module end. When the module end receives the retraining response signal, the module end enters a retraining state.

[0015] According to an embodiment of the present invention, a computer-readable storage medium is used to store a computer program, which is executed by a processor to implement the UCIe-based data transmission method.

[0016] According to an embodiment of the present invention, an electronic device includes a storage unit and a processor. The storage unit is configured to store a computer program. The processor is coupled to the storage unit and configured to execute the computer program stored in the storage unit to cause the electronic device to perform the UCIe-based data transmission method.

[0017] Based on the above, the UCIe-based data transmission method, artificial intelligence chip, medium, and electronic device of the present invention can effectively ensure the data transmission quality of the auxiliary channel between die-to-die.

[0018] In order to make the above features and advantages of the present invention more clearly understood, embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 is a schematic diagram of a data transmission system according to an embodiment of the present invention; Figure 2 Flowchart of a UCIe-based data transmission method according to an embodiment of the present invention; Figure 3 Schematic diagram of a data transmission process according to an embodiment of the present invention; Figure 4 is a flowchart of a UCIe-based data transmission method according to another embodiment of the present invention; Figure 5 is a flowchart of a UCIe-based data transmission method according to another embodiment of the present invention; Figure 6 Schematic diagram of an artificial intelligence chip according to an embodiment of the present invention.

[0020] Explanation of Figure Numbers 100: Data transmission system; 110: first core particle; 111: module end; 120: second core particle; 121: module partner end; 600: Artificial intelligence chip. DETAILED DESCRIPTION

[0021] Reference will now be made in detail to exemplary embodiments of the present invention, examples of which are illustrated in the accompanying drawings. Whenever possible, the same reference numerals are used in the drawings and the description to refer to the same or like parts.

[0022] Figure 1 Schematic diagram of a data transmission system according to an embodiment of the present invention. Figure 1 The data transmission system 100 includes a first chiplet 110 and a second chiplet 120. The first chiplet 110 and the second chiplet 120 are used to implement inter-chiplet communication based on Universal Chiplet Interconnect Express (UCIe) technology. In this embodiment, the first chiplet 110 includes a module end 111, and the second chiplet includes a module partner end 121. In this embodiment, an auxiliary channel (sideband) can be established between the module end 111 and the module partner end 121. The auxiliary channel is independent of the main data channel between the module end 111 and the module partner end 121. In this embodiment, the auxiliary channel can be composed of a clock lane (i.e., a clock signal line) and a data lane (i.e., a data signal line). In this embodiment, the first chiplet 110 and the second chiplet 120 can exchange information between the module end 111 and the module partner end 121 via sideband messages. Based on the UCIe protocol, the module partner end 121 may refer to a corresponding module end on a remote die to which the module end 111 is connected.

[0023] In this embodiment, the auxiliary channel is used to transmit control signals or management information and to ensure efficient collaboration between chiplets. For example, control commands can be transmitted between module end 111 and module partner end 121 via the auxiliary channel to help coordinate the operation of the main data channel. Chiplet status information, such as error detection or performance monitoring, can also be transmitted between module end 111 and module partner end 121 via the auxiliary channel.

[0024] In one embodiment of the present invention, the data transmission system 100 may be implemented in an electronic device, and the electronic device includes a storage unit and a processor. The storage unit is configured to store a computer program. The processor, coupled to the storage unit, is configured to execute the computer program stored in the storage unit, thereby causing the electronic device to perform the UCIe-based data transmission method described in various embodiments of the present invention.

[0025] The processor may include, for example, a central processing unit (CPU) or other programmable general-purpose or special-purpose microprocessor, a digital signal processor (DSP), a programmable controller, an application-specific integrated circuit (ASIC), a programmable logic device (PLD), other similar processing devices, or a combination of these devices.

[0026] Storage units may include, for example, random access memory (RAM), non-volatile memory, a hard disk drive (HDD), or a solid state drive (SSD). Random access memory may include, for example, dynamic random access memory (DRAM) or static random access memory (SRAM). Non-volatile memory may include, for example, flash memory or read-only memory (ROM).

[0027] Figure 2 Flowchart of a UCIe-based data transmission method according to an embodiment of the present invention. Figure 1 as well as Figure 2 , the data transmission system 100 may execute the UCIe-based data transmission method from step S210 to step S240. In step S210, the module end 111 of the first chiplet 110 communicates with the module partner end 121 of the second chiplet 120. In step S220, when the module partner end 121 receives a retraining trigger signal, the circuit of the module partner end 121 is reset, and a retraining request signal is sent to the module partner end 121 through the module end 111. In step S230, when the module partner end 121 receives the retraining request signal, a retraining response signal is replied to the module end 111 through the module partner end 121. In step S240, when the module end 111 receives the retraining response signal, the module end 111 enters a retraining state.

[0028] Specifically, when the controller of the data transmission system 100 or at least one of the first chiplet 110 and the second chiplet 120 receives a retraining trigger signal, the circuit of the module partner 121 may be reset first. Then, the module partner 111 may send a retraining request signal (e.g., including a message in the format of "SB MSG {LinkMgmt.RDI.Req Retrain}") to the module partner 121. The module partner 121 may then reply with a retraining response signal (e.g., including a message in the format of "SB MSG {LinkMgmt.RDI.Resp Retrain}") to the module partner 111, thereby causing the module partner 111 to enter a retraining state.

[0029] Matching reference Figure 3 , Figure 3 The following is a schematic diagram of the data transmission process in an embodiment of the present invention. In this embodiment, before a reset, module 111 can output a data packet to partner 121 via an auxiliary channel. However, glitches may occur on the auxiliary channel, causing partner 121 to sample a portion of the data packet incorrectly. A glitch is a brief, unexpected voltage change or pulse that occurs during signal transmission.

[0030] Therefore, in this embodiment, before the first and second chiplets 110 and 120 enter the retraining state, the module partner 121 can automatically perform a reset operation to eliminate the effects of previously sampled glitches, allowing normal sampling of transmitted data packets. Furthermore, it should be noted that reset can refer to initializing the module or clearing error states through a reset signal. The reset signal for the module partner 121 can be configured via registers. The reset signal resets the relevant receiving circuits of the module partner 121 to their initial state and clears all error states. Therefore, after the module partner 121 is reset, it can correctly sample data packets output by the module partner 111. Furthermore, the reset operation of this embodiment can also be applied to the data transmission system 100 and performed before the data transmission system 100 enters the first low-power state (L1) or the second low-power state (L2).

[0031] Figure 4 FIG. 1 is a flow chart of a UCIe-based data transmission method according to another embodiment of the present invention. Figure 1 as well as Figure 4, the data transmission system 100 may further perform the UCIe-based data transmission method as follows: Steps S401 to S412. In step S401, when entering the active state, the counter starts counting. The counter may be set in the controller, the first core 110, or the second core 120 of the data transmission system 100, but the present invention is not limited thereto. In step S402, when the counter reaches a default value, the counter stops counting, and the module end 111 sends a recalibration tracking mode request (for example, including information having the information format of "SB MSG {RECAL.track pattern init req}") to the module partner end 121. In this embodiment, when the counter reaches the default value, the counter stops counting, and in response to an auxiliary channel message parsing error, the system will trigger an interrupt and cause the counter to restart counting.

[0032] In step S403, the module partner 121 determines whether the parsing of the recalibration tracking pattern request signal is incorrect. If so, step S411 is executed to trigger an interrupt, and step S412 is executed to clear the interrupt (status) and reset the counter. If not, in step S404, the module partner 121 may send a recalibration tracking pattern response signal (e.g., including a message in the format "SB MSG {RECAL.track pattern init resp}") to the module 111. Furthermore, the module partner 121 may begin sending tracking pattern signals (track pattern) and clock pattern signals (clock pattern) to the module 111.

[0033] In step S405, module 111 determines whether the parsing of the recalibrated tracking mode response signal is incorrect. If so, step S411 is executed to trigger an interrupt, and step S412 is executed to clear the interrupt (status) and reset the counter. If not, in step S406, module 111 executes a tracking recalibration algorithm. Over extended periods of chip operation, environmental factors, chip temperature, and other factors can cause data transmission signal quality to deteriorate. This is primarily due to phase shifts between the clock and data signals. The tracking signal (track) has the same circuit structure and phase as the data signal, so recalibrating the tracking signal can improve the phase relationship between the clock and data signals. The tracking recalibration algorithm addresses this issue by recalibrating the tracking signal to retrain the phase relationship between the clock and data signals, thereby adjusting the phase relationship between the clock and data signals. In step S407, when the tracking recalibration algorithm completes execution, module 111 may send a recalibration tracking mode completion signal (e.g., including a message in the format "SB MSG {RECAL.track pattern done req}") to module partner 121. In step S408, module partner 121 may determine whether the parsing of the recalibration tracking mode completion signal is incorrect. If so, step S411 is executed to trigger an interrupt, and step S412 is executed to clear the interrupt (status) and reset the counter. If not, in step S409, module partner 121 may send a recalibration tracking mode completion confirmation signal (e.g., including a message in the format "SB MSG {RECAL.track pattern done req}") to module 111.

[0034] In step S410, module 111 determines whether the parsing of the recalibration tracking mode completion confirmation signal is incorrect. If so, step S411 is executed to trigger an interrupt, and step S412 is executed to clear the interrupt (status) and reset the counter. If not, the counter is reset to the next counting cycle.

[0035] Because glitches in the clock channel between module 111 and partner 121 can cause auxiliary channel message parsing errors, the UCIe-based data transmission method of this embodiment continuously monitors auxiliary channel signal exchange between module 111 and partner 121. Furthermore, when an interrupt is detected, the system directly clears the interrupt and jumps directly to the next round of counter timing. This effectively ensures that both module 111 and partner 121 receive correct auxiliary channel messages.

[0036] Figure 5 This is a flowchart of a UCIe-based data transmission method according to another embodiment of the present invention. The data transmission system 100 may also execute the UCIe-based data transmission method according to steps S510 to S540. In step S510, when the module end 111 enters the active state, the module partner end 121 may directly start sending the tracking mode signal and the clock mode signal to the module end 111. In step S520, the counter starts counting. In step S530, when the counter count reaches a default value, the counter stops counting, and the module end 111 may execute the tracking recalibration algorithm. In step S540, when the tracking recalibration algorithm is executed, the counter restarts the counting. In other words, the UCIe-based data transmission method of this embodiment can ensure the normal triggering of the recalibration algorithm by not adopting the auxiliary channel signal interaction method.

[0037] In addition, in one embodiment, the present invention further provides a computer-readable storage medium that can be used to store a computer program, and the computer program can be executed by a processor to implement the UCIe-based data transmission method of each of the above embodiments.

[0038] Figure 6 Schematic diagram of an artificial intelligence chip according to an embodiment of the present invention. Figure 6 In one embodiment, the artificial intelligence chip 600 may include a first core particle 110 and a second core particle 120. The first core particle 110 includes a module end 111, and the second core particle 120 includes a module partner end 121. The first core particle 110 and the second core particle 120 are used to realize inter-core communication based on universal core particle interconnection technology. The first core particle 110 and the second core particle 120 can constitute a data transmission system. In this regard, the relevant implementation methods and technical details of the first core particle 110 and the second core particle 120 can refer to the description of the above-mentioned embodiments, and the UCIe-based data transmission method between the first core particle 110 and the second core particle 120 can also refer to the processes of the above-mentioned embodiments, and sufficient guidance, suggestions and implementation instructions can be obtained. In addition, in another embodiment, the number of core particles in the artificial intelligence chip 600 is not limited to Figure 6The first core particle 110 and the second core particle 120 are shown.

[0039] In one embodiment, the artificial intelligence chip 600 can be any one of a central processing unit, a graphics processing unit (GPU), a tensor processing unit (TPU), a neural network processing unit (NPU), a deep learning processing unit (DPU), an accelerated processing unit (APU), and a general-purpose graphics processing unit (GPGPU).

[0040] In summary, the UCIe-based data transmission method, AI chip, medium, and electronic device of the present invention effectively prevent glitches on auxiliary channels (e.g., clock signal lines) between chiplets during signal transmission due to power supply noise and crosstalk between lines, which can cause errors in auxiliary channel information parsing. The UCIe-based data transmission method, AI chip, medium, and electronic device of the present invention effectively ensure that all chiplets receive correct auxiliary channel information.

[0041] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A data transmission method based on UCIe, characterized in that: include: communicating with a module partner end of a second chiplet via the module end of the first chiplet; When the module partner end receives the retraining trigger signal, resetting the circuit of the module partner end and sending a retraining request signal to the module partner end through the module end; When the module partner end receives the retraining request signal, the module partner end replies with a retraining response signal to the module end; as well as When the module side receives the retraining response signal, the module side enters a retraining state.

2. The UCIe-based data transmission method according to claim 1, wherein: Also includes: When the module terminal enters an active state, the counter starts counting; as well as When the counter reaches a default value, in response to an auxiliary channel message parsing error, an interrupt is triggered and the counter is restarted.

3. The UCIe-based data transmission method according to claim 2, wherein: Also includes: When the counter reaches the default value, sending a recalibration tracking mode request signal to the module partner end through the module end; as well as When the module partner terminal interprets the recalibration tracking mode request signal incorrectly, an interrupt is triggered by the module partner terminal, and the counter is restarted.

4. The UCIe-based data transmission method according to claim 3, wherein: Also includes: When the module partner end successfully parses the recalibration tracking mode request signal, the module partner end sends a recalibration tracking mode response signal to the module end; as well as When the module side interprets the recalibration tracking mode response signal incorrectly, an interrupt is triggered through the module side, and the counter is restarted.

5. The UCIe-based data transmission method according to claim 4, characterized in that: Also includes: When the module partner end successfully parses the recalibration tracking mode request signal, the module partner end starts to send a tracking mode signal and a clock mode signal.

6. The UCIe-based data transmission method according to claim 4, wherein: Also includes: executing a tracking recalibration algorithm by the module end when the module end successfully interprets the recalibration tracking mode response signal; as well as When the tracking recalibration algorithm is finished, a recalibration tracking mode completion signal is sent to the module partner via the module.

7. The UCIe-based data transmission method according to claim 6, characterized in that: Also includes: When the module partner terminal parses the signal of the recalibration tracking mode completion error, an interrupt is triggered by the module partner terminal, and the counter is restarted.

8. The UCIe-based data transmission method according to claim 7, characterized in that: Also includes: When the module partner end successfully parses the recalibration tracking mode completion signal, the module partner end sends a recalibration tracking mode completion confirmation signal to the module end; as well as When the module side completes the confirmation signal analysis error for the recalibration tracking mode, an interrupt is triggered through the module side, and the counter is restarted.

9. The UCIe-based data transmission method according to claim 1, wherein: Also includes: When the module end enters an active state, the module partner end directly starts sending a tracking mode signal and a clock mode signal to the module end, and causes a counter to start counting; When the counter reaches a default value, a tracking recalibration algorithm is executed by the module end; as well as When the tracking recalibration algorithm is finished, the counter is reset.

10. The UCIe-based data transmission method according to claim 2, wherein: When a glitch occurs in the clock channel between the module end and the module partner end, the auxiliary channel message parsing error occurs.

11. An artificial intelligence chip, characterized in that: include: a first core particle having a module end; as well as The second chip is coupled to the first chip and has a module partner terminal. When the module partner receives the retraining trigger signal, the circuit of the module partner is reset, and the module sends a retraining request signal to the module partner. When the module partner receives the retraining request signal, the module partner sends a retraining response signal to the module partner. When the module receives the retraining response signal, the module enters a retraining state.

12. A computer-readable storage medium for storing a computer program, characterized in that: The computer program is executed by a processor to implement the steps of the UCIe-based data transmission method according to any one of claims 1 to 10.

13. An electronic device, characterized in that: include: a storage unit for storing a computer program; as well as A processor is coupled to the storage unit and is configured to execute the computer program stored in the storage unit, so as to enable the electronic device to perform the steps of the UCIe-based data transmission method according to any one of claims 1 to 10.

Images