Skip to content
MachAI Lens Help

What Lens does

MachAI Lens is a desktop app that reads the timing reports your existing flow already produces, from Synopsys PrimeTime and Fusion Compiler, and Cadence Tempus and Innovus. Support for Tempus and Innovus is development-grade. Lens has three core workflows: Timing Review, Lens Compare and Scenario Analyzer. Digital Soul, a copilot panel, is built into each of them.

Lens reads results: it does not run timing analysis, and it does not replace PrimeTime, Fusion Compiler or your signoff flow. It is read-only with respect to your design data and never writes SDC, netlists or reports. It is not a schematic or layout viewer, and it shows no netlist or physical data. There is no clock-tree viewer, congestion map or SDC analyzer, and Digital Soul is forbidden from inventing them. Before you base a tape-out decision on anything Lens shows, verify it against the raw report.

Timing Review

Timing Review is the home screen of Lens, for triaging one run's setup and hold violations. Load your report folder and Lens indexes its scenarios and clocks for you. To set up a project, see Install Lens.

Choose your tool flow in Tool. For PrimeTime and Tempus, Scan / Detect Scenarios finds the scenarios under your root folder using your patterns, and becomes Rescan once there are scenarios. To run one scenario, pick it in Scenario:, or step to it with ◀ Prev and Next ▸, and click the ▶ button beside the selector (🗑 beside it clears that scenario's results). To run them all, click Run All Scenarios; Clear All Results clears them all. For Fusion Compiler, whose runs are per-stage files that you name individually, Scan / Detect Scenarios is not offered: pick a stage in Stage: and click Run Scenario, which runs the current stage; Clear Results clears that stage. Innovus is organised the same way as Fusion Compiler: its runs are chosen per stage, and Scan / Detect Scenarios is hidden for it too.

The four tool flows, then, come in two shapes. PrimeTime and Tempus are scenario-shaped: Lens discovers their scenarios under a run root. Fusion Compiler and Innovus are stage-shaped: you name each stage's files yourself.

Timing Review and its Timing Explorer give you:

  • per-clock WNS, TNS, violation counts, skew and latency, in a searchable, sortable table
  • the top timing paths, with a breakdown of their net segments (delay, fanout, capacitance, transition)
  • rule-based suggestions, with evidence that links back to the raw report
  • red, yellow and green status indicators that show each clock's health at a glance
  • a run health indicator (PASS, WARN or FAIL) and a warnings panel

The Timing Explorer is a focused view of one clock or one path. To open it, double-click a clock row in Per-Clock Breakdown: it opens on that clock and check, with that clock's top timing paths.

For the reports Lens reads, see Report formats.

Lens Compare

Lens Compare is for run-vs-run regression analysis across builds or stages. It gives you:

  • a ranked table of WNS and TNS regressions per scenario and clock
  • a path-diff drill-down: which paths regressed, and by how much
  • scenario matching across MMMC corners (setup/hold, PVT and mode)
  • skew insights correlated between the runs

Click Compare Runs on the Timing Review tab to open it. Its window has three modes, chosen with tab buttons: Scenario Analyzer (the default, described below), Stage Compare and Build Compare.

  • Stage Compare compares stages, for example pre-CTS, post-CTS, post-route and PT-final. It shows the per-clock change in WNS, TNS and violations, ranks the regressing clocks, and lists the paths that are common, new or removed between the baseline and the candidate.
  • Build Compare compares two builds across all their scenarios, and shows how many scenarios it matched. It accepts two PrimeTime builds or two Tempus builds: both builds must be the same tool flow, and that flow must be scenario-shaped. Fusion Compiler and Innovus organise results by stage rather than by scenario, so Build Compare refuses them and says why. To inspect the paths of one scenario, click Open Path Diff → on its row: Lens takes that scenario, clock and timing side into Stage Compare.

Comparison is same-tool only: comparing PrimeTime against Fusion Compiler, or a PrimeTime build against a Tempus build, is not offered; Lens names both tools in its refusal. For the steps, see Compare two runs.

Scenario Analyzer

Scenario Analyzer is in beta. It is a stress matrix across scenarios and clocks, for finding systemic patterns, such as one clock domain failing across all hold corners, rather than triaging path by path.

It loads a PrimeTime build's full scenario set and shows absolute margins for that single database, not deltas, and a banner says so. It is the default mode of the Lens Compare window: click Compare Runs on the Timing Review tab to reach it.

Its run picker lists each PrimeTime build as one entry, with its scenario count, and lists Fusion Compiler runs individually, one per stage. The clock-sensitivity view opens only for a PrimeTime selection. The picker does not list Tempus or Innovus builds separately; the matrix is described for PrimeTime builds.

It shows:

  • Run Selection: the run, its tool-flow tag and its PVT source line
  • Scenario Stress Overview: setup and hold WNS, setup and hold TNS, setup and hold violations, and total violations for each scenario, in a table you can sort by any column
  • Clock × Scenario Stress Matrix: a grid of scenarios against clocks, coloured red, amber or green by absolute slack
  • Derived Insights: patterns Lens detects for you, such as Systemic Bottleneck, Setup↔Hold Flip and Corner Sensitivity
  • Most Sensitive Clocks: each clock's stress percentage across the scenario set, with its worst slack and the scenario it occurred in

Click a row in the Stress Overview to select that scenario, or a cell in the matrix to select that scenario and clock, for Digital Soul's actions.

The matrix, Derived Insights and Most Sensitive Clocks need per-clock setup and hold WNS for each scenario, while the Stress Overview needs only the summary WNS, so the overview can fill in while the other three stay empty. If Lens cannot recover the per-clock evidence from the run's report, those three sections show a message that names what is missing.

Digital Soul

Digital Soul is the copilot panel built into Lens. It explains and diagnoses whatever you are looking at, such as a run, a clock, a timing path, a scenario matrix cell or a run-to-run delta, and answers with a structured card in the panel.

It is in Timing Review, the Timing Explorer, Scenario Analyzer and Lens Compare. In the main window it appears only on the Timing Review tab, once a project is open. Each view offers only the actions that suit it, and an action that needs a selection stays disabled until you make one, with a tooltip that says what to click first.

To try it:

  1. Open a project and load a scenario in Timing Review.
  2. Click the ◀ chevron on the right edge to expand the panel. It starts collapsed.
  3. Click Diagnose Current Run.
  4. Read the card. Its badges tell you how it was produced, INSPECT FIRST tells you where to look, and NEXT STEP tells you what to do.
  5. Single-click a clock row in Per-Clock Breakdown, then click Diagnose Selected Clock.

Every answer is grounded in evidence parsed from your reports, and many actions never call an LLM at all. A card that appears almost instantly comes from that deterministic path: it is rebuilt from the current evidence every time, not hard-coded or cached. If you enable generative explanations and configure an Anthropic API key, explanation actions add a richer narrative from Claude on top of that evidence. Claude is the only generative provider Lens supports. Without a key, those actions fall back to deterministic output. To add a key, see Install Lens.

The badges on a card tell you how it was produced:

  • DETERMINISTIC: built from parsed evidence and fixed analysis rules, with no LLM call
  • Claude: a narrative written by Claude
  • Deterministic fallback: a generative action that ran deterministically because the LLM was unavailable (for example, no API key) or the call failed; see Deterministic fallback
  • REVIEW REQUIRED: important evidence or validation is still unresolved
  • Hypothesis: an inference to treat as a lead, not a proven conclusion

On screen these two badges read Claude and Hypothesis, in that capitalisation. Other validation badges you may see are Verified, Engine-checked (the evidence was checked by Lens's own rules) and Limited Confidence (the evidence was insufficient or conflicting). Some older guides write CLAUDE and HYPOTHESIS in capitals; they mean the same badges.

The REASONING row switches between Quick and Extended. Extended gives Claude an extended-thinking budget and shows more actions on some views. Every result also goes into a shared history that you can open from any view. For help inside the app, click ? in the panel header.

Your reports are parsed on your machine, and the only outbound traffic is optional. Generative actions send a structured evidence pack, never raw report, log or SDC text, and deterministic actions never call out at all. See Data and privacy.

Still have a question?

Updated 19 October 2026