Secondary Foundry Design Transfer Methods for Precision Analog Circuits
Secondary foundry precision analog design transfer relies on cross-mapping PDK primitives, aligning parasitic extraction decks, and re-characterizing offset drift.

Mapping
Porting a high-precision analog integrated circuit between silicon manufacturers begins with structural netlist translation. Direct copy operations on schematic instances create functional failures when process node design kits use differing device formulations. Because precision analog topologies rely directly on physical parameters like threshold voltage, transconductance efficiency, gate-drain capacitance, and channel-length modulation, primitive devices must be mapped carefully between the primary process design kit (PDK) and the target secondary PDK.

Primitive Device Translation Rules
Transistor geometry declarations vary across process design kits. A core differential pair drawn in a 180-nanometer primary foundry process using 3.3-volt thick-oxide PMOS devices cannot map directly to an equivalent 3.3-volt node at a secondary facility without parameter conversion. Primary PDKs often specify effective gate width through finger multipliers, whereas secondary PDKs may calculate total width using a global width parameter combined with explicit finger counts.
Failing to re-evaluate width-to-length ratios against target foundry threshold voltages alters the bias current density, shifting circuit operating points into triode regions.
Passive components introduce severe translation errors. Poly-insulator-poly (PIP) capacitors common in legacy analog processes are absent in advanced secondary nodes, requiring conversion to metal-insulator-metal (MIM) structures or finger-capacitor arrays. A primary design using a PIP structure with a capacitance density of 1.5 femtofarads per square micrometer demands a physical area increase when mapped to a secondary MIM process offering only 1.0 femtofarad per square micrometer.
Poly-resistor temperature coefficients differ across foundries due to silicide block processing, implant dosage, and annealing profiles. A high-sheet-resistance poly resistor possessing a linear temperature coefficient of negative 150 parts per million per degree Celsius at the primary source can exhibit a positive 200 parts per million shift at the secondary site, disrupting bandgap voltage reference thermal compensation.
When transferring matched resistor ladders, identical physical orientation overrides identical schematic pin assignment.

Compact Model Formats across Foundry PDKs
Simulation environments rely on mathematical representations of semiconductor physics. Differences between BSIM3v3, BSIM4, and PSP (Penningroth-Smyth-Penev) compact models create discrepancies in predicted analog performance during transfer. BSIM3 models utilize empirical smoothing functions that under-predict subthreshold conduction slope, whereas PSP models deliver continuous equations from subthreshold to strong inversion.
Re-simulating a primary BSIM3-based operational amplifier under a secondary PSP PDK routinely exposes gain errors up to eight decibels in weak-inversion bias chains.
| Parameter Description | Primary PDK Parameter | Secondary PDK Parameter | Typical Mapping Discrepancy | Circuit Impact |
|---|---|---|---|---|
| Threshold Voltage Zero-Bias | VTH0 / VTO | VTH0 | 15 mV to 45 mV offset | Differential pair offset voltage drift |
| Subthreshold Swing Coefficient | NFACTOR | NFACTOR / CDEP | 5% to 12% slope shift | Sub-threshold bias current deviation |
| Mobility Temperature Exponent | UTE | UTE | 0.1 to 0.3 index delta | Transconductance shift over -40C to 125C |
| Flicker Noise Exponent | EF | EF | 0.85 versus 1.10 | Low-frequency spectral noise density shift |
| Narrow Width Effect Parameter | K3 / K3B | WL0 / W3 | Geometric scaling variation | Small-geometry current mirror mismatch |
Transistor noise parameters require individual cross-checking. The primary PDK may implement SPICE flicker noise model zero with parameters KF and AF, while the secondary PDK uses model two with frequency exponent EF. Direct numerical copying of noise parameters produces incorrect spectral noise predictions in low-noise amplifiers and continuous-time sigma-delta modulators.
Circuit designers must recalculate the total gate capacitance and parasitic substrate capacitance for every bias node before committing to layout restructuring.
When precision bandgap references drift beyond six parts per million per degree Celsius post-transfer, secondary foundries frequently cite customer layout non-compliance with proprietary well-proximity guardband guidelines.

Silicon
Physical wafer fabrication introduces local statistical variations that govern matching tolerances in differential amplifier pairs. Secondary foundry selection requires evaluating local transistor mismatch parameters, oxide charge trapping densities, and substrate isolation properties. Precision analog circuits rely heavily on physical symmetry and predictable silicon parameters to achieve low offset, high common-mode rejection, and minimal baseline drift across manufacturing lots.

Matching Coefficients and Mismatch Scaling
Pelgrom equations express threshold voltage mismatch variance as a function of effective gate area. The mismatch parameter AVth defines the process-dependent matching capability of a specific foundry node. A primary foundry process achieving an AVth of 3.5 millivolt-micrometers requires a secondary process with an identical or lower coefficient to maintain input offset voltage specifications without enlarging transistor gate dimensions.
- Threshold Voltage Offset Drift Transistor threshold mismatch between paired differential devices generates systematic input offset voltages that shift output baselines across temperature gradients.
- Flicker Noise Floor Elevation Interfacial oxide traps at secondary manufacturing facilities alter 1/f noise corner frequencies, degrading SNR in precision signal paths.
- Capacitance Ratio Non-Linearity Dielectric thickness variations across MIM layers introduce voltage-dependent capacitance variations that distort switched-capacitor converter stages.
- Substrate Parasitic Injection Inadequate well isolation profiles cause ground noise cross-talk between high-speed digital logic blocks and microvolt analog inputs.
Device scaling during porting cannot assume linear relationships for matching. Increasing gate area reduces statistical mismatch variance, but increases parasitic gate-to-substrate capacitance. When porting a 16-bit analog-to-digital converter, doubling transistor width to compensate for a higher secondary foundry AVth parameter increases capacitive loading at critical nodes, lowering bandwidth and operational amplifier phase margin.

Low Frequency Noise and Substrate Noise Isolation
Trapping mechanisms at the gate oxide interface dictate pink noise spectral density in deep sub-micron process nodes. The normalized flicker noise parameter KF varies dramatically depending on thermal anneal protocols and dielectric deposition methods used by the secondary factory. PMOS devices fabricated with buried-channel designs exhibit lower 1/f noise than surface-channel alternatives.
Porting a low-noise amplifier from a surface-channel primary process to a secondary foundry utilizing different nitridation steps requires physical re-measurement of test structures to verify noise performance.
Substrate noise coupling degrades analog precision when mixed-signal blocks occupy the same die. Deep N-well structures mitigate carrier injection into the bulk p-type substrate. The effectiveness of deep N-well isolation depends on guard ring layout, pickup tap spacing, and substrate resistivity.
Primary foundries utilizing high-resistivity substrates above twenty ohm-centimeters provide greater low-frequency isolation than secondary foundries operating on standard ten ohm-centimeter wafers. Adding triple-well guard bands and substrate contact rings prevents ground bounce from modulating analog reference nodes.
Inadequate guard ring spacing during porting leads to substrate cross-talk that degrades signal-to-noise ratio by twelve decibels, forfeiting two effective bits in high-resolution converter designs.

Parasitics
Interconnect impedance and cross-coupling capacitance fundamentally alter the frequency response of operational amplifiers. Moving an analog layout to a secondary manufacturing site invalidates previous parasitic extraction results due to variations in metal thickness, inter-metal dielectric constants, and trench isolation geometry. Precision designs require comprehensive parasitic extraction deck validation to avoid unexpected parasitic poles and stability degradation.

Layout Extraction Rule Deck Conversion
Electronic design automation toolchains calculate physical line resistance using foundry-supplied technology files. Interconnect layer stacks differ between semiconductor manufacturers even at identical lithographic nodes. A primary node employing four metal layers with a thick top power metal of two micrometers may transfer to a secondary facility offering five metal layers with a maximum top metal thickness of 0.8 micrometers.
This thickness difference increases power distribution trace resistance and shifts local ground potential under transient load conditions.
| Extraction Category | Primary Foundry Deck | Secondary Foundry Deck | Target Tolerance | Verification Method |
|---|---|---|---|---|
| Metal-to-Metal Fringe Capacitance | Field-Solver Derived | Empirical Lookup Table | Under 3% total delta | 3D Field Solver Cross-Check |
| Substrate Coupling Capacitance | Included in PEX Deck | Requires Explicit Deep N-Well Model | Under 5% total delta | Test Structure S-Parameter Test |
| Poly Interconnect Resistance | Silicide Block Aware | Global Sheet Resistance Average | Under 2% total delta | 4-Point Kelvin Probe Measurement |
| Via Array Series Resistance | Per-Via Enclosure Aware | Lumped Via Resistance Value | Under 4% total delta | Via-Chain Test Structure Measurement |
Extraction tools must re-calculate net capacitance based on the secondary foundry dielectric constants. Fluorinated silicate glass dielectrics yield a relative permittivity of approximately 3.5, whereas standard silicon dioxide yields 3.9. Parasitic extraction decks that neglect sidewall coupling capacitance under-predict total capacitance on tightly routed differential signal traces.
A second-source fabrication contract specifying undefined extraction deck rules leaves all substrate capacitance re-characterization expenses with the silicon purchaser.

Stress Dependent Physical Effects in Precision Analog Layouts
Mechanical compressive forces from shallow trench isolation shift PMOS carrier mobility across active diffusion regions. Length of diffusion (LOD) and well proximity effect (WPE) parameters alter transistor drain currents based on distance from isolation edges and retrograded well implants. Secondary PDK rule decks must incorporate stress-dependent parameters (SC, SBA, SBC) within SPICE netlists during physical extraction.
Placing critical matched transistors closer than three micrometers to a shallow trench isolation boundary causes asymmetric mobility shifts up to four percent. Layout modification strategies require extending dummy oxide diffusion zones around precision current mirrors. Standard differential layouts must maintain uniform dummy gate structures to balance compressive stress vectors across matched gate channels.
Engineering teams continue to debate whether 3D field solvers eliminate the need for silicon iteration when transferring fifty-femtofarad capacitor ladders between disparate process nodes.

Corner
Process extreme definitions establish the simulation boundaries used to verify circuit performance across manufacturing spreads. Primary and secondary foundries calculate process skews using different statistical criteria. A fast-fast or slow-slow designation from one manufacturer does not represent identical parameter deviations at a second facility.

How Do Process PDK Corner Models Diverge between Primary and Secondary Foundries?
Foundry PDK releases structure simulation skew files using distinct statistical distribution boundaries. Primary foundries often specify process corners at three-sigma limits from mean electrical test results. Secondary facilities may define skew models using non-centrosymmetric six-sigma bounding boxes or worst-case power-speed trade-off envelopes.
Transistor threshold voltage shifts combined with gate oxide thickness tolerances alter transconductance calculations across extreme temperature ranges. A circuit designed to operate at 125 degrees Celsius under slow-slow process conditions at a primary site may experience excessive speed degradation or bias shift when evaluated under secondary slow-slow models. Re-centering bias networks ensures that current sources maintain saturation across all secondary process combinations.

Monte Carlo Yield Prediction Reconciliation
Statistical yield analysis requires aligned variance models to predict circuit offset distributions. Mismatch models rely on Monte Carlo analysis to simulate device parameter variations across thousands of run iterations. Secondary foundries sometimes omit spatial correlation parameters in standard Monte Carlo distributions, leading to over-optimistic differential matching estimations.
- Extract netlists from primary layout using target foundry deck parameters.
- Run five-hundred-point Monte Carlo sweeps across industrial temperature ranges.
- Compare offset distributions against primary foundry Monte Carlo baselines.
- Adjust dummy transistor guard bands until mismatch variance falls within target limits.
Evaluating global lot-to-lot variation alongside local device-to-device mismatch ensures reliable yield forecasting. When primary Monte Carlo data indicates a three-sigma offset distribution of 1.2 millivolts, but secondary simulation predicts 2.1 millivolts, layout engineers must enlarge active channel areas to return offset bounds to baseline specifications.
The standard silicon procurement agreement specifies that wafer lot acceptance relies strictly on PCM test structure parameters rather than functional chip yield across process extremes.

Mask
Photolithographic reticle generation represents the final physical boundary between layout design and physical lithography. Transferring CAD data between foundries demands exact layer translation, DRC deck conversion, and alignment of optical proximity correction rules.

Stream Format Conversion and Layer Number Alignment
Data interchange formats carry physical geometry designations between CAD software and mask writing equipment. GDSII and OASIS file structures map physical drawn shapes to specific numerical layer numbers and datatype designations. Primary and secondary factories almost never utilize identical layer mapping conventions.
- Layer Map Definition File Translates primary drawn layout numbers to secondary process reticle designations without altering polygon coordinates.
- LVS Rule Translation Script Aligns schematic netlist device pins with secondary physical layer structures to prevent connectivity verification errors.
- PEX Resistance Extraction Table Maps physical metal layer codes to specific sheet resistance equations within target foundry deck file structures.
- GDSII Stream File Signoff Validates mathematical equivalence between primary and secondary target geometry definitions before reticle generation.
A pin geometry drawn on layer 31 with datatype 0 in a primary PDK might represent deep N-well, whereas layer 31 datatype 0 in a secondary PDK designates a silicide-block region. Taping out without converting layer mapping tables results in missing process steps and fabricated silicon failure.
A three-phase shift in optical proximity correction rules increases input stage offset voltage by 1.8 millivolts when sub-micron STI boundaries sit closer than two micrometers from differential channel gates.

Physical Verification Decks and Optical Proximity Correction
Lithography correction algorithms modify drawn polygons to compensate for optical diffraction phenomena. Secondary foundries run proprietary optical proximity correction (OPC) software after design signoff. Layout features that pass physical design rules (DRC) can fail during OPC processing if drawn feature spacing creates lithographic conflicts.
Precision analog designs require specific keep-out zones to protect sensitive matched structures from OPC modifications. Adding OPC keep-out layers prevents mask generation software from adding aggressive hammerhead or serif shapes near differential pair gates, maintaining spatial symmetry across critical matching boundaries.
A design transfer package without verified GDSII layer translation tables forces the fabrication plant to assume default layer mappings that risk catastrophic short circuits.

Custody
Commercial arrangements govern intellectual property ownership, source file deliverables, and financial risk allocation during secondary facility bring-up. Establishing clear technical boundaries within contract statements of work prevents ownership disputes when porting analog blocks.

Non Recurring Engineering Amortization and Reticle Costs
Non-recurring engineering expenses combine mask set fabrication, engineering runs, and laboratory re-characterization. Mask sets for sub-micron nodes represent significant capital commitments. Allocating risk between buyer and secondary foundry requires explicit agreement regarding wafer acceptance standards and minimum yield thresholds.
Unbundling non-recurring engineering costs from per-unit production prices exposes true manufacturing economics. Amortizing NRE costs over small production runs inflates component unit cost, while large volumes justify initial engineering expenditures.

Worked Dual Source Re Characterization Scope and Budget
Consider a precision operational amplifier porting program moving from a primary 180-nanometer BCD process to a secondary foundry facility. The engineering team must re-qualify a low-drift, high-voltage amplifier servicing industrial sensor interfaces. The budget allocation covers schematic translation, PDK deck alignment, layout modifications, mask production, engineering prototype runs, and bench characterization.
The primary baseline operational amplifier occupies 1.2 square millimeters of silicon, operates up to 36 volts, and achieves an input offset voltage under 50 microvolts with a temperature drift below 0.5 microvolts per degree Celsius. The secondary foundry PDK features similar 36-volt high-voltage LDMOS and thick-oxide PMOS devices, but exhibits a higher AVth mismatch parameter (4.2 millivolt-micrometers versus 3.2 millivolt-micrometers at the primary site) and different poly-resistor temperature coefficients.
| Transfer Phase | Engineering Hours | NRE / Deliverable Cost (USD) | Schedule Duration | Primary Risk / Verification Target |
|---|---|---|---|---|
| PDK Rule & Netlist Mapping | 120 Hours | 18,000 | 3 Weeks | Device parameter translation and SPICE model alignment |
| Layout Re-Design & Rescaling | 280 Hours | 42,000 | 6 Weeks | LOD/WPE stress mitigation and capacitor ladder matching |
| DRC / LVS / PEX Signoff | 90 Hours | 13,500 | 2 Weeks | 3D parasitic extraction and DRC deck verification |
| Mask Reticle Set Fabrication | 0 Hours | 110,000 | 5 Weeks | Full mask set generation for 180nm BCD node |
| MPW / Engineering Wafer Run | 40 Hours | 35,000 | 8 Weeks | Silicon fabrication of 25 prototype engineering samples |
| Bench Bring-Up & AEC-Q100 Test | 160 Hours | 28,000 | 4 Weeks | Offset voltage drift and noise floor characterization |
| Total Program Scope | 690 Hours | 246,500 | 28 Weeks | Complete dual-sourcing transfer verification |
Engineering labor is priced at an average internal rate of 150 USD per hour across 690 total project hours, yielding an engineering labor expenditure of 103,500 USD. Adding direct external mask and wafer costs of 145,000 USD brings total non-recurring expenses to 248,500 USD (including initial test board fabrication).
To compensate for the secondary foundry higher AVth matching coefficient, layout engineers increase the input differential transistor pair channel area from 100 square micrometers to 175 square micrometers. This physical resizing increases input stage gate capacitance from 180 femtofarads to 315 femtofarads, lowering the amplifier unity-gain bandwidth from 10 megahertz to 8.2 megahertz unless tail bias current rises by 120 microamperes.
Foundry process transfer timelines expand by eighty business days when third-party analog IP blocks require independent re-characterization.
Recalculating power distribution networks shows that the additional current draw increases baseline supply power by 0.43 milliwatts per channel. Incorporating these adjustments during the layout phase prevents a secondary engineering wafer run, saving 35,000 USD in prototype wafer expenses and eight weeks of schedule delay.
Engineering teams finalize the handover dossier by locking the revision hashes of all netlists, extraction scripts, and test benches in a joint escrow repository.





