On-chip Monitor Analytics Scales With Silicon Chip Production From NPI Through HVM

By Guy Cortez and Dan Alexandrescu

At the New Product Introduction (NPI) stage of silicon chip production, product engineers work with a limited but critical dataset – typically from initial silicon samples or engineering lots – enabling early assessment of the power and performance of your silicon. Analytics solutions typically have no time-to-results (TTR) issues when the volume of data is manageable, as is the case during NPI. However, when production of the silicon goes into High Volume Manufacturing (HVM), which commonly entails millions of chip wafer production, analytics solutions may not be able to manage this high volume of data, preventing you from continuing to monitor key operational metrics throughout production. Moreover, there are also important results obtained during NPI that should be leveraged and help prepare for HVM. We will discuss in this blog how the Monitor Analytics solution from Synopsys can help when transitioning to high volume production.

This is the third blog from Synopsys written on monitor analytics. The first blog introduced the concept of how analytics on data originating from embedded on-chip monitors can be automated and leveraged to measure key operational metrics such as power and performance of your silicon chip, resulting in improved engineering productivity and time-to-market. The second blog focused on a specific use case of Monitor Analytics called Silicon Model Calibration, which can aid in producing better designs with lower margins by leveraging real-world measurements of on-chip monitors from production silicon.

This third and latest blog highlights the need for distinct Monitor Analytics solutions as data volume increases, as well as addresses requirements from different persona types, such as designers and product engineers, as your chip goes through the various stages of manufacturing production.

Typically, design teams, once they have taped out their design and it goes on to production, quickly move on to the next design project, while product engineering teams are tasked with ensuring the manufactured design meets the requisite specifications as the design becomes a silicon device. Design-related issues that may arise during chip production, potentially many months later after the design team is well onto their next project, may still require support from design teams to address any problems. This gap in time makes the troubleshooting of issues nontrivial, and in a worst-case scenario, the designer of that taped-out design may no longer be working at the company. Having a solution that can quickly analyze the state of your design and pinpoint the root cause of your post-silicon problems immediately as silicon becomes available significantly mitigates such risks.

While having a Monitor Analytics solution together with gaining immediate access to monitor-rich data in the early New Product Introduction (NPI) stage in manufacturing is critical to quickly finding and correcting design-related issues, it is equally important to maintain an analytics solution throughout the remaining High-Volume Manufacturing (HVM) stage. However, the types of analysis that are performed in NPI vary when going into HVM, along with the associated careabouts of designers and product engineers. Therefore, the requirements in an analytics solution differs from NPI to HVM.

An analytics solution tailored for designers and NPI specialists

The Monitor Analytics NPI solution, as depicted in Figure 1, is used during the engineering phase to analyze small lots of devices in detail. The solution focuses on data exploration, gap-to-target analysis and in-depth silicon to model correlation. Monitor Analytics NPI is flexible and can adapt to various devices from highly-specialized test vehicles to complex commercial products. Test results can be provided in a variety of formats, such as STDF, CSV, Test Logs, and Unified Data Format (UDF). UDF, designed by Synopsys, serves as the standard format for data exchange within the Silicon Lifecycle Management (SLM) platform. UDF specifications are available to Synopsys customers on demand.

Fig. 1: Monitor Analytics solution targeted for NPI.

The NPI specialist uses Monitor Analytics NPI for in-depth analysis of a limited set of results from the initial runs. They carefully interpret the results of SLM IPs, other monitors, DFT, and BIST over a wide range of test conditions and algorithms. They will interact strongly with the designers and IP owners that often look for high-quality silicon data for their monitors. They understand what is important and what is not when interpreting the results.

Key analytics available early with NPI results:

Parametric Sensitivity Characterization will measure how parameters (e.g., RO Frequency, path delay, leakage, gain) vary with temperature and voltage. Process Corner and Environmental Comparison verify if the device meets specs under full operating range over the full process space. The test results can also provide an understanding of how behavior at one condition predicts behavior at others.

Design of Experiments (DOE) Analysis will provide insight into how process or design variables (e.g., transistor size, layout, standard cell library parameters) impact performance.

The learnings gleamed from the NPI results will later guide adaptive test limits or binning strategies, will prioritize high-risk corners for inline screening or quality monitoring and will enable test skipping or predictive test modeling as your product moves into high-volume manufacturing.

As an example of an NPI to HVM handoff, NPI-derived sigma thresholds (e.g. -3σ or -6σ) can be tentatively set as initial test limits during high volume manufacturing. The NPI engineer can use the Monitor Analytics NPI solution to check if the conditions are met: NPI lot is representative of the final production conditions, measurement noise is well understood and the NPI test conditions are covering the HVM needs.

In Table I below, the statistics of the path slack measurements performed during NPI are evaluated to calculate test limits applicable to HVM, while Figure 2 shows the actual test results during HVM.

Table I. NPI-driven HVM limits

Fig. 2: NPI-driven limits applied during HVM.

The NPI specialist prepares the collaterals (including MA configurations – UDF) and works with the volume specialist to quickly enable a Monitor Analytics cloud-based solution with the NPI collaterals.

While an NPI solution may be a file-based approach and gated only by the memory of the local server or laptop used by the user, an HVM solution needs to rely on a scalable database architecture commonly seen in cloud computing, such as a Kubernetes cluster of servers, in order to support high volumes of data. The servers used are meant to be scalable, and the environment can grow with additional servers as the demand for higher volumes of data increases. Even though this type of database architecture is associated with popular 3^rd party cloud providers such as Microsoft’s Azure or Amazon’s AWS, users can optionally deploy their own cloud-like infrastructure onsite on their work premises (aka on-prem) with the support from the analytics provider if they choose to host this data themselves.

An analytics solution tailored for product engineers and HVM specialists

HVM test results provide rich statistical and spatial data that go far beyond simple pass/fail metrics. Product Engineers/Volume specialists will benefit from high quality, efficient volume analytics for the SLM monitors. Furthermore, Monitor Analytics Cloud can extract specific, actionable insights that guide yield improvement, quality control and test optimization as shown in Figure 3.

Fig. 3: Monitor Analytics solution targeted for HVM.

The HVM data analysis pipeline is typically optimized for data size and throughput. Monitor Analytics Cloud will use the data already loaded in the Synopsys Silicon.da data analytics platform. While the NPI format input flexibility is lost, the space and time cardinality of the data will provide greater opportunity for analytics.

Spatial Monitor Measurements Patterns (at Wafer and Die Level) provide an insight into intra-die variation, localized hot spots or systematic layout effects. Monitor Analytics Cloud provides semantics-based analysis of monitor test results and, through the statistics-based guided navigation, presents problematic areas first to the data scientist, enabling a fast understanding of any process issue.

Temporal trends, such as Process Drift, are more easily observable in the thousands of wafers from the different sub-lots and lots. Monitor Analytics Cloud presents parametric univariate and multi-variate trend reports, highlighting monitor configurations that are susceptible to show early signs of process drift. The following chart in Figure 4 presents the results of one of the monitors that shows significant lot-to-lot drift. In this example, the later lot shows results that are slower in performance than the expected slow corner simulations, which could result in timing failures.

Fig. 4: Lot to lot variation captured and analyzed across typical, slow and fast PMOS/NMOS process corners.

Furthermore, the HVM data analyst can pinpoint critical issues both in space and time and export a focused snapshot back to the Monitor Analytics NPI. This can be triggered by unusual patterns or clustered yield dips, when new corner cases or process splits are introduced, or when a suspected quality or reliability issue emerges. As an example, the following wafer map view in Figure 5 is produced by Monitor Analytics NPI using selective data from HVM after normalization to the expected simulation targets.

Fig. 5: Selective subset of HVM data captured in Monitor Analytics NPI solution.

Future use case considerations such as Aging and Degradation monitoring could be implemented. Monitor Analytics can play a foundational role in modeling device aging using HTOL (High Temperature Operating Life) test results and preparing aging models fit for In-Field analytics. Monitor Analytics could also leverage this to perform health assessment and root cause analysis for monitor results that are reported periodically or sporadically from in-field.

Benefits and conclusions

By establishing a structured collaboration framework, NPI and Volume data analysts can use Monitor Analytics to:

• Accelerate time-to-yield by transferring knowledge from NPI to HVM
• Reduce false alarms and manual rule tuning
• Enable predictive quality monitoring with a shared model foundation

The synergetic use of Monitor Analytics NPI and Cloud solutions aligns teams across the product lifecycle, ensuring that insights built on a few hundred devices can guide decisions across millions.

Dan Alexandrescu is a principal R&D engineer at Synopsys.