Simulation method for effectively solving non-convergence problem of DDR simulation waveform
By dividing the frequency range into important energy bands and high-frequency energy decreasing bands, and adopting logarithmically spaced frequency point settings, the non-convergence problem caused by unreasonable frequency point settings in DDR signal simulation is solved, and efficient simulation result convergence and accuracy are achieved.
Patent Information
- Application Number
- CN202510637488.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-16
- Publication Date
- 2025-09-23
AI Technical Summary
In the existing technology, DDR signal simulation suffers from the problem of non-convergence due to unreasonable frequency point settings, especially in high-frequency, wide-band, and multi-port scenarios, where the calculation complexity is high and the accuracy is insufficient.
The frequency range is divided into important energy frequency bands within the fundamental frequency range and high-frequency energy decreasing frequency bands. A logarithmic frequency point setting strategy is adopted in the high-frequency energy decreasing frequency band to reduce the density step by step. Combined with dense fundamental frequency sampling, the number of frequency points is reduced.
The convergence optimization of DDR signal simulation results is achieved while maintaining the accuracy of key parameters, reducing the computational complexity and the number of frequency points, and avoiding the numerical calculation divergence caused by oversampling in traditional methods.
Smart Images
Figure CN120688416A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of DDR simulation, and in particular to a simulation method for effectively solving the problem of non-convergence of DDR simulation waveforms. Background Art
[0002] With the rapid development of DDR technology, DDR signal rates have increased from several hundred Mbps in the early days to several Gbps today, and the number of storage chips integrated into systems has also continued to increase in line with the demand for high-density design. Under this trend, DDR signal integrity simulation faces two core challenges. First, the increase in signal rate has led to an expansion of the frequency range that needs to be extracted. For example, a 1.6Gbps signal needs to cover 0-5GHz to meet the 3x frequency industry standard. Second, the number of system ports has increased dramatically. For example, the number of ports in a single link can reach hundreds. As a result, when dealing with high-frequency wide-band and multi-port scenarios, traditional simulation methods often cause "simulation non-convergence" problems due to unreasonable frequency point setting strategies. That is, the waveform diverges and the amplitude increases abnormally in the time domain simulation, seriously affecting R&D efficiency and the accuracy of signal integrity assessment.
[0003] The existing technology mainly adopts the default adaptive frequency sampling strategy, such as linear uniform sampling or the software built-in Adaptive mode. Its core logic is to evenly distribute the frequency points within the target frequency range or automatically allocate the sampling density based on preset rules. For example, for 1.6Gbps DDR signals, the industry's conventional practice is to extract the 0-5GHz frequency range, set the frequency points in a linear or adaptive manner, and ultimately generate about 350 frequency point data. It can be seen that the total number of frequency points is too large, and when the number of ports reaches more than 200, the combined calculation amount of frequency points and ports increases exponentially, which eventually leads to numerical calculation instability and causes the waveform to not converge. Figure 2 .
[0004] If the frequency points of the entire frequency band are simply reduced to reduce the amount of calculation, key energy information will be lost due to insufficient sampling, resulting in signal amplitude and phase distortion and insufficient accuracy; if dense sampling of the high-frequency band is retained to maintain accuracy, the problem of non-convergence cannot be avoided.
[0005] The above problems are worth solving. Summary of the Invention
[0006] In order to overcome the problems of high computational complexity and simulation non-convergence caused by high-frequency oversampling in the prior art, as well as the inability to balance computational efficiency while ensuring accuracy, the present invention provides a simulation method that effectively solves the problem of non-convergence of DDR simulation waveforms.
[0007] The technical solution of the present invention is as follows:
[0008] A simulation method for effectively solving the problem of non-convergence of DDR simulation waveforms includes the following steps:
[0009] Determine the base frequency based on the operating rate of the DDR signal;
[0010] The frequency range of S parameter extraction is divided into an important energy frequency band within the fundamental frequency range and a high-frequency energy decreasing frequency band outside the fundamental frequency range;
[0011] A linearly spaced frequency setting strategy is adopted for the important energy frequency band, and the high-frequency energy decreasing frequency band is divided into at least two sub-frequency bands according to the energy distribution characteristics. A logarithmically spaced frequency setting strategy is adopted for each sub-frequency band, and the logarithmic spacing density of different sub-frequency bands decreases step by step with increasing frequency;
[0012] The S parameters extracted using the above frequency point setting strategy are used to perform full-channel waveform simulation.
[0013] As a preferred technical solution of the present invention, the base frequency is half of the DDR signal operating rate.
[0014] As a preferred technical solution of the present invention, the important energy frequency band includes the range from the DC point to the fundamental frequency, and the high-frequency energy decreasing frequency band includes the range from the fundamental frequency to at least 3 times the fundamental frequency.
[0015] Furthermore, the important energy frequency bands include:
[0016] Independently set DC point to capture signal DC bias characteristics;
[0017] The fundamental frequency linear sweep frequency band has a starting frequency that is the non-zero low frequency point after removing the DC point, and an ending frequency that is the fundamental frequency.
[0018] As a preferred technical solution of the present invention, the linearly spaced frequency setting strategy is: setting the frequencies at fixed frequency intervals, and the fixed frequency intervals are 5 to 15 MHz.
[0019] As a preferred technical solution of the present invention, the high-frequency energy decreasing frequency band at least includes:
[0020] The first sub-band has a range from the fundamental frequency to twice the fundamental frequency, and its frequency points are set using the first logarithmic interval density;
[0021] The second sub-band has a range from 2 times the fundamental frequency to 3 times the fundamental frequency, and its frequency point setting adopts the second logarithmic interval density;
[0022] And the second logarithmic interval density is smaller than the first logarithmic interval density.
[0023] Furthermore, the first logarithmic interval density is 10 to 25 points per decade, and the second logarithmic interval density is 5 to 15 points per decade.
[0024] As a preferred technical solution of the present invention, the logarithmically spaced frequency setting strategy includes:
[0025] In each sub-band, the frequency ratio of adjacent frequency points satisfies: f i+1 =f i ×10 1 / n ; Among them, f i is the frequency of the ith frequency point, and n is the logarithmic interval density value corresponding to the sub-band.
[0026] As a preferred technical solution of the present invention, a verification step is also included, including:
[0027] Obtain a signal waveform from the waveform obtained by using the uniform frequency setting method, recorded as waveform A, and then obtain a signal waveform from the result obtained in step 4, recorded as waveform B, and verify the convergence of waveform B.
[0028] Furthermore, the verification step also includes accuracy retention verification, the specific steps are:
[0029] In the data interval where waveform A converges, compare the voltage difference, waveform rising edge time deviation, and waveform falling edge time deviation between waveform A and waveform B.
[0030] The present invention according to the above scheme has the following beneficial effects:
[0031] The present invention divides the frequency range into an important energy band within the fundamental frequency range and a high-frequency energy decreasing band outside the fundamental frequency range. The high-frequency energy decreasing band is further subdivided into at least two sub-bands according to the energy distribution characteristics. Different sub-bands are set with logarithmic intervals, and the logarithmic interval density decreases step by step with increasing frequency, which significantly reduces the number of frequency points in the high-frequency band, fundamentally reduces the computational complexity, and avoids the problem of numerical calculation divergence caused by "oversampling".
[0032] Moreover, by adopting a strategy of gradually decreasing the logarithmic interval density with increasing frequency in the high-frequency energy decreasing frequency band, and coordinating with the dense sampling of the fundamental frequency, the simulation results are almost identical to the waveform of the traditional full-band dense sampling. While reducing the number of frequency points by 64%, the accuracy of key parameters such as amplitude and edge is not affected. Therefore, the present invention achieves convergence optimization while taking into account accuracy, breaking the technical bottleneck of "either no convergence or sacrificing efficiency" in the traditional method. BRIEF DESCRIPTION OF THE DRAWINGS
[0033] Figure 1 is a flow chart of the method of the present invention;
[0034] Figure 2 The multi-signal waveform diagram obtained by simulating the existing technology does not converge the waveform;
[0035] Figure 3This is a multi-signal waveform diagram obtained by simulation of the present invention;
[0036] Figure 4 A waveform comparison diagram of a signal is selected for the present invention and the prior art;
[0037] Figure 5 The figure is a comparison chart of the simulation accuracy of the present invention and the prior art in the waveform convergence range of the prior art. DETAILED DESCRIPTION
[0038] To better understand the objectives, technical solutions, and technical effects of the present invention, the present invention is further explained below with reference to the accompanying drawings and embodiments. It should be noted that similar reference numerals and letters represent similar items in the following drawings. Therefore, once an item is defined in one drawing, it does not need to be further defined or explained in subsequent drawings. It should also be noted that the embodiments described below are intended only to illustrate the present invention and are not intended to limit the present invention.
[0039] It should be noted that the indicated orientation or positional relationship is based on the orientation or positional relationship shown in the accompanying drawings, or the orientation or positional relationship in which the product of the application is typically placed when in use, or the orientation or positional relationship commonly understood by those skilled in the art, or the orientation or positional relationship in which the product of the application is typically placed when in use. This is merely for the convenience of describing this application and simplifying the description, and does not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operate in a specific orientation. Therefore, it should not be understood as a limitation on this application. The terms "first" and "second" are used only for the convenience of description and should not be understood to indicate or imply relative importance or implicitly indicate the number of technical features.
[0040] like Figure 1 As shown, a simulation method for effectively solving the problem of non-convergence of DDR simulation waveform includes the following steps:
[0041] Step 1: Determine the base frequency according to the operating rate of the DDR signal;
[0042] Specifically, read the signal operating rate from the DDR chip specification or system design document. For example, if the rate is 1.6 Gbps, it is recorded as f data Calculate the base frequency: According to signal processing theory, the base frequency is half of the DDR signal operating rate, that is, the base frequency f base =f data / 2, for example, 1.6Gbps corresponds to a base frequency of 800MHz.
[0043] Step 2: Divide the frequency range of S parameter extraction into an important energy frequency band within the fundamental frequency range and a high-frequency energy decreasing frequency band outside the fundamental frequency range;
[0044] Specifically, most of the signal's energy falls within the fundamental frequency range. Therefore, the frequency band within the fundamental frequency range is regarded as an important energy band. Therefore, with the fundamental frequency as the core, the frequency band from the DC point (0Hz) to the fundamental frequency range is defined, such as 0Hz to 800Hz. The frequency band outside the fundamental frequency range is defined as a high-frequency energy decreasing frequency band. The high-frequency energy decreasing frequency band can be the fundamental frequency to 3 times the fundamental frequency, such as the 1st frequency range of 800MHz to 1.6GHz, the 2nd frequency range of 1.6GHz to 3.2GHz, and the 3rd frequency range of 3.2GHz to 4.8GHz. The high-frequency energy decreasing frequency band can also be the fundamental frequency to 4 times the fundamental frequency, or the fundamental frequency to 5 times the fundamental frequency, etc.
[0045] Step 3: A linearly spaced frequency point setting strategy is adopted for the important energy frequency band. Specifically, the important energy frequency band includes an independently set DC point and a base frequency linear scanning frequency band; wherein the DC point (0 Hz) is used as a frequency point alone to process the DC bias characteristics of the signal; the base frequency linear scanning frequency band starts from the non-DC low frequency band to the base frequency, that is, the starting frequency is the non-zero low frequency point after removing the DC point, and the ending frequency is the base frequency, and linearly spaced dense sampling is adopted.
[0046] The linear frequency setting strategy involves setting frequencies at fixed intervals between 5 and 15 MHz, including 5MHz, 6MHz, 7MHz, 8MHz, 9MHz, 10MHz, 11MHz, 12MHz, 13MHz, 14MHz, and 15MHz. To use this strategy, select "Custom Frequency Range" in the simulation software's frequency setting interface. Set the starting frequency of the fundamental frequency linear sweep range to 10MHz and the ending frequency to 800MHz. Set the "Linear Sweep" mode to a fixed 10MHz step size. The software will then automatically generate a sequence of 80 frequency points within the critical energy band: 10MHz, 20MHz, 30MHz, and so on to 800MHz.
[0047] It can be seen that the important energy frequency band physically includes the DC point (0Hz) and the fundamental frequency linear sampling frequency band, but the starting frequency of the linear interval setting is the first valid frequency point after removing the DC point. This first valid frequency point is the value of "0 + fixed frequency interval", rather than linear sampling starting from 0Hz. 0Hz is a separate DC point and does not participate in linear scanning.
[0048] Step 3 also includes: dividing the high-frequency energy decreasing frequency band into at least two sub-frequency bands according to the energy distribution characteristics, each sub-frequency band adopts a logarithmic frequency setting strategy, and the logarithmic spacing density of different sub-frequency bands decreases step by step as the frequency increases. Specifically, the high-frequency energy decreasing frequency band includes:
[0049] The first sub-band has a range from the fundamental frequency to twice the fundamental frequency, and its frequency points are set using the first logarithmic interval density;
[0050] The second sub-band has a range from 2 times the fundamental frequency to 3 times the fundamental frequency, and its frequency point setting adopts the second logarithmic interval density;
[0051] Similarly, a third sub-band, a fourth sub-band, and so on may also be included; and the logarithmic interval density of the latter sub-band is smaller than the logarithmic interval density of the former sub-band.
[0052] In a specific embodiment, the high-frequency energy decreasing frequency band includes a 1-time frequency sub-band, a 2-time frequency sub-band, and a 3-time frequency sub-band; wherein the 1-time frequency sub-band ranges from the base frequency to 2 times the base frequency, such as 800MHz to 1.6GHz; the 2-time frequency sub-band ranges from 2 times the base frequency to 3 times the base frequency, such as 1.6GHz to 3.2GHz; and the 3-time frequency sub-band ranges from 2 times the base frequency to 3 times the base frequency, such as 3.2GHz to 5GHz. The operating steps for frequency scanning setting are as follows:
[0053] For the 1 octave sub-band, select "Logarithmic Sweep" mode, set the start frequency to 800 MHz, the end frequency to 1.6 GHz, and the interval step size to 20 points per decade. In other alternative embodiments, the first logarithmic interval density for the first sub-band can be 10, 15, or 25 points per decade, etc.
[0054] For the 2-octave sub-band, select "Logarithmic Sweep" mode, set the start frequency to 1.6 GHz, the end frequency to 3.2 GHz, and the interval step size to 10 points per decade. In other alternative embodiments, the second logarithmic interval density for the second sub-band can also be 10 or 15 points per decade, as long as it is less than the first logarithmic interval density for the first sub-band.
[0055] For the 3rd octave sub-band, select "Logarithmic Sweep" mode, set the start frequency to 3.2 GHz, the end frequency to 5 GHz, and the interval step size to 5 points per decade. In other alternative embodiments, the third logarithmic interval density of the third sub-band can also be 3, 8, or 10 points per decade, as long as it is less than the second logarithmic interval density of the second sub-band.
[0056] It should be noted that each decade refers to the interval where the frequency range is expanded by 10 times. The frequency change within each decade is logarithmic. For example, the frequency span from 10MHz to 100MHz is a decade. The frequency points are rounded according to engineering practices. In each sub-band, the frequency ratio of adjacent frequency points satisfies: f i+1 =f i ×10 1 / n ; Among them, f i is the frequency of the ith frequency point, and n is the logarithmic interval density value corresponding to the sub-band.
[0057] By entering the start frequency, end frequency, and logarithmic spacing density for each sub-band, the high-frequency band setting is completed. The software automatically summarizes the frequency points of all frequency bands, reducing the total number of points to 124 frequency points, a 64% reduction compared to the original 350 frequency points.
[0058] Step 4. Use the S parameters extracted by the above frequency point setting strategy to perform full channel waveform simulation; specifically, import the PCB / layout file, define the signal network and ports, such as 299 ports, apply the above frequency setting strategy, start the S parameter simulation, and the software automatically calculates and simulates the signal waveforms of all ports to ensure that there is no non-convergence. Figure 3 , the figure is a waveform diagram with time (ns) as the horizontal axis and voltage (V) as the vertical axis.
[0059] The method of the present invention further comprises:
[0060] Step 5: Verification step, specifically including the following steps:
[0061] Step 501: Obtain a comparison waveform;
[0062] For a clearer comparison, a signal waveform is obtained from the waveform obtained by the traditional method using the uniform frequency setting method, which is recorded as waveform A. Then, a signal waveform is obtained from the result obtained in step 4, which is recorded as waveform B.
[0063] Step 502: Convergence verification;
[0064] Reference Figure 4 , observe the stability of waveforms A and B within the full simulation time (0 to 80ns); it can be seen from the figure that the divergent area of waveform A presents a stable signal shape in waveform B;
[0065] Step 503: Verify accuracy retention;
[0066] In the data range where waveform A can converge (0 to 28ns), compare the signal characteristics of waveform A and waveform B, refer to Figure 5 It is found that the absolute value of the voltage difference between waveform B and waveform A at the corresponding time points (0 to 28ns) is very small, and the rising edge / falling edge time deviation is very small. The two waveforms almost overlap. Therefore, it is determined that the signal simulation accuracy is not reduced after reducing the number of points by adopting the method of the present invention.
[0067] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0068] The above embodiments merely illustrate several implementations of the present invention, and while their descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent. It should be noted that a person skilled in the art would be able to make numerous variations and improvements without departing from the spirit of the present invention, all of which fall within the scope of protection of the present invention. Therefore, the scope of protection of the patent for this invention shall be determined by the appended claims.
Claims
1. A simulation method that effectively solves the problem of non-convergence of DDR simulation waveform, characterized in that: The following steps are involved: Determine the base frequency based on the operating rate of the DDR signal; The frequency range of S parameter extraction is divided into an important energy frequency band within the fundamental frequency range and a high-frequency energy decreasing frequency band outside the fundamental frequency range; A linearly spaced frequency setting strategy is adopted for the important energy frequency band, and the high-frequency energy decreasing frequency band is divided into at least two sub-frequency bands according to the energy distribution characteristics. A logarithmically spaced frequency setting strategy is adopted for each sub-frequency band, and the logarithmic spacing density of different sub-frequency bands decreases step by step with increasing frequency; The S parameters extracted using the above frequency point setting strategy are used to perform full-channel waveform simulation.
2. The simulation method for effectively solving the non-convergence of DDR simulation waveform according to claim 1, characterized in that: The base frequency is half of the DDR signal operating rate.
3. The simulation method for effectively solving the non-convergence of DDR simulation waveform according to claim 1, characterized in that: The important energy frequency band includes the range from the DC point to the fundamental frequency, and the high-frequency energy decreasing frequency band includes the range from the fundamental frequency to at least 3 times the fundamental frequency.
4. The simulation method for effectively solving the non-convergence of DDR simulation waveform according to claim 3, characterized in that: The important energy frequency bands include: Independently set DC point to capture signal DC bias characteristics; The fundamental frequency linear sweep frequency band has a starting frequency that is the non-zero low frequency point after removing the DC point, and an ending frequency that is the fundamental frequency.
5. The simulation method for effectively solving the non-convergence of DDR simulation waveform according to claim 1, characterized in that: The linear frequency setting strategy is to set the frequencies at fixed frequency intervals, where the fixed frequency intervals are 5 to 15 MHz.
6. The simulation method for effectively solving the non-convergence of DDR simulation waveform according to claim 1, characterized in that: The high-frequency energy decreasing frequency band at least includes: The first sub-band has a range from the fundamental frequency to twice the fundamental frequency, and its frequency points are set using the first logarithmic interval density; The second sub-band has a range from 2 times the fundamental frequency to 3 times the fundamental frequency, and its frequency point setting adopts the second logarithmic interval density; And the second logarithmic interval density is smaller than the first logarithmic interval density.
7. The simulation method for effectively solving the non-convergence of DDR simulation waveform according to claim 6, characterized in that: The first logarithmically spaced density is 10 to 25 points per decade, and the second logarithmically spaced density is 5 to 15 points per decade.
8. The simulation method for effectively solving the non-convergence of DDR simulation waveform according to claim 1, characterized in that: The logarithmically spaced frequency setting strategy includes: In each sub-band, the frequency ratio of adjacent frequency points satisfies: f i+1 =f i ×10 1 / n ; Among them, f i is the frequency of the ith frequency point, and n is the logarithmic interval density value corresponding to the sub-band.
9. The simulation method for effectively solving the non-convergence of DDR simulation waveform according to claim 1, characterized in that: Also included are verification steps, including: Obtain a signal waveform from the waveform obtained by using the uniform frequency setting method, recorded as waveform A, and then obtain a signal waveform from the result obtained in step 4, recorded as waveform B, and verify the convergence of waveform B.
10. The simulation method for effectively solving the non-convergence of DDR simulation waveform according to claim 9, characterized in that: The verification step also includes accuracy retention verification, the specific steps are: In the data interval where waveform A converges, compare the voltage difference, waveform rising edge time deviation, and waveform falling edge time deviation between waveform A and waveform B.