Get email delivery of the Cadence blog featured here
EMIR analysis is one of the most challenging fields of circuit simulation. It requires the power and/or signal net parasitics to be preserved for the IR drop and EM current analysis that is performed later. At the same time, the EMIR analysis requires SPICE-like accuracy to check the EM currents against the current limits defined by the foundry. Voltus-Fi XL, the Cadence transistor-level EMIR analysis tool, uses the Spectre EMIR solution for simulating the circuit behavior and calculating the IR drop and EM current data. In this blog, we will explore the benefits of Spectre X in EMIR analysis and also compare its performance with Spectre APS for both direct and iterated EMIR methods.
Spectre X EMIR Use Model
The Spectre X EMIR use model is similar to the Spectre APS EMIR use model. You just need to define the required simulation preset and enable the EMIR flow using the +emir command-line option.
The EMIR flow automatically disables parasitic optimization and preserves all the parasitic elements for the IR and EM analysis.
Spectre X EMIR Accuracy Validation
Before deploying Spectre X in EMIR simulation, we need to check what accuracy it provides in comparison to our established golden Spectre APS EMIR accuracy. Only if the Spectre X solver provides acceptable EMIR accuracy, we can look at its performance and capacity benefits.
For comparing the Spectre X EMIR accuracy with the Spectre APS EMIR accuracy, we selected a large 28nm memory design with 1.2M MOSFETs, 40M resistors, and 20M capacitors, and compared the equivalent Spectre APS/X accuracy modes in the direct EMIR method. Each graph in the figure below shows the RMS EM current comparison for all power net resistors between the given Spectre APS/X mode (y-axis), and the golden accuracy reference, Spectre APS, with the ++aps=conservative mode (x-axis).
From the above EM RMS current comparison, we can observe that the Spectre X AX preset shows similar accuracy as its equivalent Spectre APS +aps=moderate (+mod) mode. We can also see that there is a similar degradation in accuracy when moving to higher performance modes. We can therefore conclude that Spectre X with an equivalent preset provides similar EMIR accuracy as Spectre APS.
Spectre X Direct EMIR Method Performance
The Spectre direct EMIR method simulates the entire design, including all parasitic R and C elements. It poses one of the most challenging simulation requirements for Spectre. For analyzing the benefit of Spectre X direct method over Spectre APS, we used the same large 28nm memory design with 1.2M MOSFETs, 40M resistors, and 20M capacitors.
For this design, the power net IR drop and EM analysis in the Spectre APS ++aps=liberal mode on 8 cores took 2d 17h, while Spectre X VX took 1d 7h, which is a 2.1x performance gain. We saw in the previous section that Spectre X provides similar accuracy as Spectre APS for this mode. Besides the performance gains, Spectre X also reduced the overall memory consumption. For this case, Spectre APS used 260GB, while Spectre X used 230GB.
Spectre X provides a new technology for high core multi-threaded and distributed simulation. When we run the same simulation using Spectre X on a machine using 32 cores (using the Spectre command-line option +mt=32), we get an additional 2x speed gain and the simulation time is reduced to about 9h. There is no accuracy degradation when moving from 8 cores to 32 cores.
1.2M MOS, 40M R, 20M C
From the above, we can conclude that Spectre X provides the following benefits over Spectre APS when using the direct EMIR method:
We have already discussed in the Spectre Tech Tips: Spectre X Update blog that Spectre X provides more value for large and complex designs, especially for advanced node designs.
Spectre X Iterated EMIR Method Performance
The Spectre iterated EMIR method consists of the following two stages:
The Spectre APS or Spectre X solver is used in the first stage, while the second stage uses a specialized RC network solver. Therefore, the Spectre X performance gain on the overall iterated method EMIR analysis depends on the time that is spent in each stage of the EMIR analysis. For example, there will be no performance gain for a given design if the first stage takes 30min and the second stage runs for 24h. On the other hand, if the first stage is dominating, or at least occupying a significant portion of the overall time, Spectre X may provide a performance gain.
For evaluating the Spectre X performance impact on the iterated method, we used a large SRAM design and compared the performance of the Spectre X MX preset with the Spectre APS ++aps=moderate mode. For both the simulators, 8 cores were used and the accuracy was confirmed to be similar.
7M N, 600k MOS, 40MC, 7M R
From the above, we can conclude that for such design and EMIR scenario, the Spectre X solver provides significant performance gain even in the iterated method. Generally, the Spectre X performance gain in the iterated method is expected to be less compared to the direct method; however, in terms of absolute simulation time, we can still expect the iterated method to be much faster.
Introducing Spectre X in EMIR Analysis
About Spectre Tech Tips
Spectre Tech Tips is a blog series aimed at exploring the capabilities and potential of Spectre®. In addition to providing insight into the useful features and enhancements in Spectre, this series broadcasts the voice of different bloggers and experts, who share their knowledge and experience on all things related to Spectre. Enter your email address in the Subscriptions box and click SUBSCRIBE NOW to receive notifications about our latest Spectre Tech Tips posts.