Method for adjusting a communication bandwidth within a protocol stack of a chiplet

Abstract

Method (100) for adjusting a communication bandwidth within a protocol stack of a chiplet or on a chip between different parts of the chip, which comprises a plurality of communication channels, comprising the steps of: providing (110) a total bandwidth of the system, detecting (120) a number of the communication channels and their physically possible communication bandwidth, detecting (140) a communication signal by means of a protocol, detecting (150) a priority of the communication signal by means of a protocol layer of the protocol stack, and adjusting (160) the communication bandwidth for the communication signal as a function of the total bandwidth, the number of communication channels and their communication bandwidth, and the priority of the communication signal.

Description

[0001] R. 419590

[0002] - 1 -

[0003] Description

[0004] title

[0005] Method for setting a communication bandwidth within a protocol stack of a chiplet

[0006] The invention relates to a method for setting a communication bandwidth within a protocol stack of a chiplet.

[0007] State of the art

[0008] Die-to-die connections play a crucial role in chiplet ecosystems. The UCIe standard, which comprises three layers—protocol layer, die-to-die adapter, and physical layer—is currently being developed for this purpose.

[0009] The protocol stack is implemented as hardware IP on the chip and can cover all three layers. The protocol layer is integrated into an existing chip infrastructure using a standard protocol, such as AXI. For this purpose, the protocol adapter converts a protocol specified in the UCIe standard, such as FDI, to the chip's internal protocol variant, for example, AXI.

[0010] The die-to-die adapter ensures error correction via forward error correction (FEC) or reverse error correction to achieve a minimum bit and packet error rate, e.g., BER < 10⁻¹⁰. 15 error / bit, to ensure this minimum error rate is prescribed in the UCIe standard.

[0011] The physical layer is now configured for a dedicated speed grade (e.g., 16, 24, 32, 64 GT / s) in a chiplet system. The fixed speed grade influences, for example, the transmission quality, which can be measured by the bit error rate or signal-to-noise ratio, etc. The preset speed also results in a possible Rp. 419590

[0012] - 2 -

[0013] The maximum transmission line length between chiplets through the substrate is limited. This leads to restrictions in the design of a chiplet system. The chosen substrate material also influences the transmission quality by affecting the characteristic impedance, signal attenuation, and transmission power. The substrate cannot be considered constant in terms of material and structure across different chiplet system configurations.

[0014] The disadvantage here is that the fixed speed and the associated bandwidth limitation do not allow for flexibility in prioritizing the processing of security-critical information.

[0015] Furthermore, when variable speeds are used, for example via the UCIe standard, it is not always guaranteed that the full communication bandwidth can be provided. This can lead to problems with the necessary data exchange with other participants in some applications that require a specific bandwidth.

[0016] In addition, modern automotive systems have further requirements that go beyond ferry operation. For example, a so-called sentry mode detects objects passing by when the vehicle is stationary and, if necessary, saves this information to persistent storage. Power consumption is also an important criterion here and must be minimized.

[0017] The purpose of the invention is to overcome these disadvantages.

[0018] Disclosure of the invention

[0019] The inventive method for setting a communication bandwidth within a protocol stack of a chiplet having a plurality of communication channels comprises providing a total bandwidth of the chiplet, acquiring a number of the communication channels and their electrically possible communication bandwidth, acquiring a communication channel by means of a protocol, and acquiring a priority of the communication signal by means of a protocol layer of the protocol stack. Furthermore, R. 419590

[0020] - 3 -

[0021] Methods include setting the communication bandwidth for the communication signal depending on the required total bandwidth, taking into account the application states, and deriving a priority for communication data streams.

[0022] The advantage here is that the bandwidth is dynamically adjustable, which, for example, has an advantage for the power loss of the chiplet and its aging.

[0023] In a training course, the possible transmission speed of the respective communication channels is determined, and the communication bandwidth for the communication signal is assessed depending on possible application modes (e.g.,

[0024] Parking, Driving, Sentry, Entertainment) discontinued.

[0025] The advantage here is that, despite variable transmission speeds, sufficient bandwidth is available for the respective application in the application mode taking place at the operating point.

[0026] In a further version, the protocol includes an AXI protocol.

[0027] The advantage here is that AXI is an industry-wide standard for connecting different hardware IPs.

[0028] In a further education course, the protocol stack is based on a chip-to-chip transmission standard, in particular the UCIe standard.

[0029] The advantage here is that the UCIe standard enables cross-manufacturer communication compatibility.

[0030] Further advantages arise from the following description of exemplary embodiments or the dependent patent claims.

[0031] Brief description of the drawings R. 419590

[0032] - 4 -

[0033] The present invention is explained below with reference to preferred embodiments and the accompanying drawings. These show:

[0034] Figure 1 shows a method for setting a communication bandwidth within a protocol stack of a chiplet.

[0035] Figure 1 shows the method 100 for setting a communication bandwidth within a protocol stack of a chiplet that has a multitude of communication channels. The method 100 starts by providing 110 a necessary total bandwidth of the chiplet derived from an application mode. In a subsequent step 120, the number of communication channels and their electrically possible communication bandwidth are determined. In a subsequent step 140, a communication signal is detected using a protocol. In a subsequent step 150, the priority of the communication signal is determined using a protocol layer of the protocol stack. In a subsequent step 160, the communication bandwidth for the communication signal is set depending on the total bandwidth, the number of communication channels and their respective communication bandwidths, and the priority of the communication signal.

[0036] In one embodiment, non-safety-critical information is available for transmission, e.g., from the infotainment system, so-called IVI data. This data has a lower priority compared to safety-critical information, e.g., data of category ASIL-D. If safety-critical information is available for transmission simultaneously with the non-safety-critical information, the bandwidth is dynamically allocated according to the priority of the available information. The non-safety-critical information is allocated less communication bandwidth on the interface than the safety-critical information. An application example is the simultaneous transmission of ADAS information and IVI information. A very large communication bandwidth is allocated to the ADAS information ad hoc, while the IVI information is transmitted using the remaining available bandwidth and may be suspended. R. 419590

[0037] - 5 -

[0038] If only non-safety-critical information is present, it can be transmitted using the full communication bandwidth. However, as soon as safety-critical information is present at the interface, the transmission of non-safety-critical data must sacrifice bandwidth.

[0039] In another embodiment, different application modes can be considered during prioritization. For example, ADAS information is frequently used in artificial intelligence, including in conjunction with image data. However, these image sources are in different driving situations, i.e., driving, parking, sentry mode, possibly other camera sources, and / or different resolutions / frame rates. This affects the number of pixels to be processed, for example, by an attached AI chiplet. Therefore, at lower resolutions / frame rates, the die-to-die connection can be operated more slowly to, for example, save energy.

[0040] Another advantageous application is maintaining ferry operations at a lower frame rate, allowing the vehicle to continue driving autonomously, albeit at a reduced speed. Other similar applications are conceivable, all sharing the common goal of reducing the function's performance while maintaining safe ferry operation, and then later, after the business trip or vacation, taking the vehicle to the workshop.

[0041] Since modern processes result in significantly smaller structures, the effects of factors such as defects, HCl, TDDB, or electromigration are considerable and lead to high field failure rates over long operating times. The methods described above help to avoid generating failures and instead only cause limitations.

[0042] In another embodiment, high-priority information is assigned a low static bandwidth to allow for dynamic and flexible responses. The bandwidth remains constant for short messages.

[0043] In another embodiment, low-priority information has a low static bandwidth. R. 419590

[0044] - 6 -

[0045] In a further training course, in step 130, which is carried out between step 120 and step 140, the transmission speed of the respective communication channels is recorded and the communication bandwidth for the communication signal is additionally set depending on the transmission speed.

[0046] In one embodiment, low-priority information can be transmitted in other grids, i.e., at a slower speed and with lower bandwidth.

[0047] In another embodiment, high-priority information can be transmitted in the grid with the highest speed and bandwidth.

[0048] The protocol can be, for example, an AXI protocol, and the protocol stack can be based on, for example, the UCIe standard.

[0049] The described implementation examples can be implemented via Quality of Service. Two use cases can be optimized in this way. Firstly, UCIe with variable frequency and the dependence on the system state, i.e., ADAS AI requires more bandwidth for transmission than the I Vl components, so their bandwidth is then reduced.

[0050] In other words, if only a certain bandwidth is available, the communication bandwidth in systems with multiple communication channels is divided across an interface. As is generally accepted practice, one can proceed statically and assign a specific share to each application, but this has the disadvantage of providing less bandwidth for individual applications than if the allocation is done dynamically depending on the system state, as proposed here.

Claims

R. 419590 - 7 - Claims 1. Method (100) for setting a communication bandwidth within a protocol stack of a chiplet or on a chip between different parts of the chip having a plurality of communication channels, comprising the steps: • Providing (110) a total system bandwidth, • Identifying (120) a number of communication channels and their physically possible communication bandwidth, • Capture (140) a communication signal using a protocol, • Capturing (150) a priority of the communication signal by means of a protocol layer of the protocol stack, • Setting (160) the communication bandwidth for the communication signal depending on the total bandwidth, the number of communication channels and their communication bandwidth, and the priority of the communication signal.

2. Method (100) according to claim 1, characterized in that a transmission speed of the respective communication channels is detected and the communication bandwidth for the communication signal is set depending on the transmission speed.

3. Method (100) according to one of claims 1 or 2, characterized in that the protocol comprises an AXI protocol.

4. Method (100) according to one of the preceding claims, characterized in that the protocol stack is based on a chip-to-chip transmission standard.

Images