jitx-interconnect-constraints

JITX Interconnect Constraints

Apply signal integrity constraints to signal topologies in JITX Python designs. This skill bridges jitx-circuit-builder (wiring with +) and jitx-substrate-modeler (routing structures) by teaching how to create SI-aware topologies with >> and constrain them.

Environment

Environment setup is handled by the base jitx skill. Ensure it has been invoked first.

Package Architecture

# Core SI imports
from jitx.si import (
    Constrain,
    ConstrainDiffPair,
    ConstrainReferenceDifference,
    DiffPairConstraint,
    SignalConstraint,
    PinModel,
    BridgingPinModel,
    TerminatingPinModel,
    ReferencePlanes,
    RoutingStructure,
    DifferentialRoutingStructure,
    symmetric_routing_layers,
)

# Topology and ports
from jitx.net import DiffPair, Port, TopologyNet, Topology

# Common bundles
from jitx.common import LanePair

# Core framework
from jitx import Circuit, Net, current
from jitx.toleranced import Toleranced
from jitx.units import ohm

These DO NOT EXIST — never import: jitx.topology, jitx.constraints.si, jitx.diffpair, jitx.signal, jitx.si.TopologyNet (it is in jitx.net), jitxlib.si, jitxlib.constraints

Key locations:

• DiffPair and TopologyNet are in jitx.net
• Topology is in jitx.net (re-exported from jitx.si)
• All constraint classes are in jitx.si
• Routing structures are in jitx.si (also used in substrate definitions)

For complete API signatures, see jitx.si API docs.

Anti-string-hacking — read before applying per-lane / per-pair constraint metadata

When applying constraints across N parallel diff-pairs or N parallel lanes, iterate over a typed collection (self.lanes: list[DiffPair]), not a string-keyed dict (self.pairs["TX_b0"]) or sibling attributes plus getattr. The constraints themselves are objects; assemble them inside a typed dataclass or list, never dict[str, Any]. See jitx/references/architectural-patterns.md §§ "String-keyed dicts → structural objects" and "Sibling attributes → array attributes" before writing per-lane constraint application code.

For a same-model self-critique pass on the constraint code after writing (catches what these rules don't), invoke jitx-code-review. Optional for single-task use.

The Critical Distinction: + vs >>

The most important concept for SI-constrained designs:

# + creates a Net — electrical connection only, NO SI awareness
# The routing engine can route this any way it wants
self.power_net = self.ic.VCC + self.cap.p1

# >> creates a TopologyNet — ordered signal path WITH SI awareness
# The routing engine knows the signal order and can apply constraints
self += self.driver.out >> self.receiver.inp

Storage rules (same as for + nets):

# CORRECT — store on self with +=
self += self.src >> self.dst

# CORRECT — store as named attribute
self.my_topo = self.src >> self.dst

# CORRECT — store in a list on self
self.topos = [self.src >> self.dst]

# WRONG — topology created but not stored, silently dropped!
self.src >> self.dst  # BAD — lost!

When to use which:

• + — Power rails, ground connections, control signals without SI requirements
• >> — Any signal with timing, skew, loss, or impedance constraints (clocks, data buses, differential pairs, RF signals)

Basic Topology and Constraints

Step 1: Create topology with >>

# >> defines the ordered signal path
self += self.driver.out >> self.receiver.inp

Step 2: Identify topology for constraint application

# Topology(begin, end) is a DESCRIPTOR — it identifies the path, does not create it
topo = Topology(self.driver.out, self.receiver.inp)

Step 3: Apply constraints

# Get routing structure from substrate (defined in jitx-substrate-modeler)
rs50 = current.substrate.routing_structure(50.0)

# Apply with ReferencePlanes context
self.GND = Net(name="GND")
with ReferencePlanes(self.GND):
    self.cst = Constrain(topo).insertion_loss(1.0).structure(rs50)

Constraint methods on Constrain

# Insertion loss — max dB
Constrain(topo).insertion_loss(3.0)

# Timing — max propagation delay (seconds)
Constrain(topo).timing(500e-12)

# Timing — min and max delay window
Constrain(topo).timing(high=500e-12, low=100e-12)

# Routing structure assignment
Constrain(topo).structure(rs50)

# Chaining (all constraints applied together)
self.cst = (
    Constrain(topo)
    .insertion_loss(1.0)
    .timing(500e-12)
    .structure(rs50)
)

Multiple signals with same constraint

# Constrain accepts a sequence of Topology objects
topos = [Topology(s, d) for s, d in zip(src_pins, dst_pins)]
self.cst = Constrain(topos).insertion_loss(3.0).structure(rs50)

Differential Pairs

The DiffPair bundle

import jitx
from jitx.net import DiffPair

class MyComponent(jitx.Component):
    data = DiffPair()  # has .p (positive) and .n (negative) sub-ports

Constraining differential pairs

# Create diff pair topology — >> connects both .p and .n
self += self.src.data >> self.dst.data

# Identify and constrain
topo = Topology(self.src.data, self.dst.data)

drs100 = current.substrate.differential_routing_structure(100.0)

self.dp_cst = (
    ConstrainDiffPair(topo)
    .timing_difference(5e-12)   # max 5ps intra-pair skew (P-to-N)
    .insertion_loss(3.0)         # max loss per line
    .structure(drs100)           # 100 ohm differential routing structure
)

DiffPairConstraint — reusable helper

For applying the same constraint to multiple differential pairs:

from jitx.si import DiffPairConstraint

# Create once, use many times
dp_cst = DiffPairConstraint(
    skew=Toleranced(0, 5e-12),    # max 5ps intra-pair skew
    loss=3.0,                      # max 3dB insertion loss
    structure=drs100,              # routing structure
)

# Apply to each diff pair
self += self.src.tx >> self.dst.rx
dp_cst.constrain(self.src.tx, self.dst.rx)

self += self.src.clk >> self.dst.clk
dp_cst.constrain(self.src.clk, self.dst.clk)

LanePair — TX + RX bundle

from jitx.common import LanePair

class MySerialPort(Port):
    lane = LanePair()  # has .TX (DiffPair) and .RX (DiffPair)

Bus-Level Matching (ConstrainReferenceDifference)

Match multiple data signals to a reference clock or strobe:

# Create topologies for clock and data
self += self.src.clk >> self.dst.clk
clk_topo = Topology(self.src.clk, self.dst.clk)

data_topos = []
for i in range(8):
    self += self.src.data[i] >> self.dst.data[i]
    data_topos.append(Topology(self.src.data[i], self.dst.data[i]))

# Constrain all data signals relative to clock
# "All data signals must arrive within +/- 11ps of the clock"
self.bus_skew = (
    ConstrainReferenceDifference(
        guide=clk_topo,             # reference signal
        topologies=data_topos,      # signals to match
    ).timing_difference(Toleranced(0, 11e-12))
)

Key concept: The guide is the reference signal. All topologies are constrained to arrive within the specified timing window relative to the guide.

This pattern is used for:

• RGMII: data-to-clock matching
• DDR: DQ-to-DQS matching, CMD/ADDR-to-CK matching
• PCIe/SFP: inter-lane skew (first lane as reference)

Pin Models

Pin models characterize signal propagation through components for SI analysis.

TerminatingPinModel — IC endpoints

Applied to active component pins where the signal terminates:

class MyIC(jitx.Component):
    TX_P = Port()
    TX_N = Port()

    # Silicon-to-pad delay and loss
    pm_txp = TerminatingPinModel(TX_P, delay=15e-12, loss=0.2)
    pm_txn = TerminatingPinModel(TX_N, delay=15e-12, loss=0.2)

BridgingPinModel — pass-through components

Applied to passive components that a signal passes through (AC coupling caps, series resistors, common-mode chokes):

class BlockingCapacitor(jitx.Component):
    p1 = Port()
    p2 = Port()

    # Signal passes through with this delay and loss
    pin_model = BridgingPinModel(p1, p2, delay=6e-12, loss=0.5)

Using BridgingPinModel in a subcircuit

Define a subcircuit with ports that a topology chains through:

class ACCoupler(Circuit):
    """Single-ended AC coupling subcircuit."""
    A = Port()
    B = Port()

    def __init__(self):
        self.cap = BlockingCapacitor()  # has BridgingPinModel
        self += self.A >> self.cap.p1
        self += self.cap.p2 >> self.B

Differential pair version:

class DiffPairCoupler(Circuit):
    """AC coupling for a differential pair."""
    A = DiffPair()
    B = DiffPair()

    def __init__(self, capacitance=100e-9):
        self.cap_p = BlockingCapacitor(capacitance)
        self.cap_n = BlockingCapacitor(capacitance)

        # Topology chains through the caps
        self.topo_p1 = self.A.p >> self.cap_p.p1
        self.topo_p2 = self.cap_p.p2 >> self.B.p
        self.topo_n1 = self.A.n >> self.cap_n.p1
        self.topo_n2 = self.cap_n.p2 >> self.B.n

Chaining topology through a subcircuit

The outer circuit connects to the subcircuit's ports with >>. The topology chains through the subcircuit's internal >> connections automatically:

class MyDesign(Circuit):
    def __init__(self):
        self.driver = MyDriver()
        self.recv = MyReceiver()
        self.coupler = ACCoupler()

        # Topology chains: driver -> coupler.A -> [internal] -> coupler.B -> recv
        self += self.driver.out >> self.coupler.A
        self += self.coupler.B >> self.recv.inp

        # Constrain the full end-to-end path
        topo = Topology(self.driver.out, self.recv.inp)
        with ReferencePlanes(self.GND):
            self.cst = Constrain(topo).insertion_loss(3.0)

Topology with bridging components (no subcircuit)

When a topology passes through a component that does not have an embedded BridgingPinModel, define one at the circuit level:

# Signal path: driver → cap → receiver
self += self.driver.out >> self.cap.p1
self += self.cap.p2 >> self.receiver.inp

# Add BridgingPinModel so constraint engine tracks delay/loss through the cap
self.bridge = BridgingPinModel(self.cap.p1, self.cap.p2, delay=6e-12, loss=0.5)

# Constraint sees the full path including cap delay/loss
topo = Topology(self.driver.out, self.receiver.inp)
self.cst = Constrain(topo).insertion_loss(3.0)

Reference Planes

Reference planes specify which net serves as the return path for each routing layer.

Context manager form (most common)

self.GND = Net(name="GND")

# All reference layers use GND
with ReferencePlanes(self.GND):
    self.cst = Constrain(topo).structure(rs50)

Per-layer form

# Different reference nets per layer
with ReferencePlanes({0: self.GND, 1: self.GND, 2: self.PWR}):
    self.cst = Constrain(topo).structure(rs50)

On Constrain directly (via structure ref_layers)

self.cst = Constrain(topo).structure(rs50, ref_layers={0: self.GND})

Important: If a routing structure has .reference() definitions (from substrate-modeler), you MUST provide ReferencePlanes. Without them, the constraint will error at build time.

Tag-based routing structures (alternative to Constrain)

Since 4.2, a routing structure can also be applied through the design-rule system: tag the nets/routes, then attach the structure as a rule effect.

from jitx.constraints import Tag, design_constraint

class HighSpeedTag(Tag):
    "Nets that route on the 50-ohm structure."

# in the circuit — tag the nets (see jitx-physical-layout "Layout-intent tags"):
HighSpeedTag().assign(self.clk_net)

# in the rules — attach the structure:
self.hs_rule = design_constraint(HighSpeedTag()).routing_structure(rs50, ref_net=self.GND)

When to choose which:

• Topology Constrain(...).structure(...) — an ordered >> path exists and SI constraints (timing, skew, insertion loss) travel with the structure. The structure applies to the topology paths you enumerate (Constrain takes one topology or a list — see "Multiple signals with same constraint").
• Tag-based design_constraint(...).routing_structure(...) — class-of-net rules: every net or code-based Route carrying the tag gets the structure, with no per-signal topology authoring. This is also the only way to put a structure on a plain Net or a code-based Route, neither of which has a >> topology.

Reference planes for the rule effect resolve through ref_net= (one net for all reference layers), ref_layer_nets={layer: net}, or an active ReferencePlanes context — note the kwarg names differ from Constrain.structure(..., ref_layers=). Full signature and the three resolution modes: jitx-substrate-modeler "Routing structures as a rule effect".

Building Protocol Constraints (SignalConstraint)

For reusable protocol-specific constraint bundles, subclass SignalConstraint[T]:

Pattern: Port bundle + Standard + Constraint

from dataclasses import dataclass
from jitx.si import SignalConstraint, DiffPairConstraint
from jitx.net import DiffPair, Port
from jitx.toleranced import Toleranced
from jitx import current

# 1. Define port bundle
class MySerialLink(Port):
    tx = DiffPair()
    rx = DiffPair()

# 2. Define standard parameters
@dataclass(frozen=True)
class MyStandard:
    skew: Toleranced = Toleranced(0, 1e-12)
    loss: float = 12.0
    impedance: Toleranced = Toleranced(100, 10)

# 3. Define constraint class
class MyConstraint(SignalConstraint["MySerialLink"]):
    def __init__(self, standard: MyStandard,
                 structure: DifferentialRoutingStructure | None = None):
        if structure is None:
            structure = current.substrate.differential_routing_structure(
                standard.impedance
            )
        self._dp_cst = DiffPairConstraint(
            skew=standard.skew, loss=standard.loss, structure=structure
        )

    def constrain(self, src: MySerialLink, dst: MySerialLink):
        # Crossover: TX→RX, RX→TX
        self._dp_cst.constrain(src.tx, dst.rx)
        self._dp_cst.constrain(dst.tx, src.rx)

Using constrain_topology (context manager)

The constrain_topology method handles topology creation and constraint application together. Use it when inserting components (like AC coupling caps) into the topology:

class MyDesignCircuit(Circuit):
    def __init__(self):
        self.fpga = FPGA()
        self.phy = PHY()

        std = MyStandard()
        cst = MyConstraint(std)

        # constrain_topology creates proxies and applies constraints
        with cst.constrain_topology(
            self.fpga.serial, self.phy.serial
        ) as (src, dst):
            # Insert AC coupling caps in the TX path
            self.coupler = DiffPairCoupler()
            self.topos = [
                src.tx >> self.coupler.A,
                self.coupler.B >> dst.rx,
                dst.tx >> src.rx,  # RX path direct
            ]

Hierarchical constraints (DDR4 pattern)

For complex protocols with multiple signal groups, compose constraints:

class DDR4Constraint(SignalConstraint["DDR4"]):
    def __init__(self, standard):
        self._data_cst = DDR4DataConstraint(standard)
        self._acc_cst = DDR4AccConstraint(standard)

    def constrain(self, src: DDR4, dst: DDR4):
        self._data_cst.constrain(src.data, dst.data)
        self._acc_cst.constrain(src.acc, dst.acc)
        # Cross-group: CK-to-DQS timing
        ...

See jitx_protocols_ext for complete protocol implementations including PCIe, SATA, SFP, DDR4, LPDDR4, LPDDR5, and GDDR7.

Working with Built-in Protocols

JITX provides built-in protocol constraints in jitxlib. Check your installed version for available protocols with your Grep tool (pattern class.*Constraint.*SignalConstraint, path .venv, glob *.py); it recurses and is OS-agnostic. Shell fallback: bash grep -rn "class.*Constraint.*SignalConstraint" .venv/lib/python*/site-packages/jitxlib/protocols/ (macOS/Linux); on Windows use the Grep tool, or Select-String over .venv\Lib\site-packages\jitxlib\protocols.

Common built-in protocols:

Protocol	Import Path	Port Bundle
USB	jitxlib.protocols.usb	USB2, USB3
DisplayPort	jitxlib.protocols.displayport	DisplayPort
RGMII	jitxlib.protocols.ethernet.mii.rgmii	RGMII
100Base-TX	jitxlib.protocols.ethernet.mdi.mdi100base_tx	MDI100BaseTX
1000Base-T	jitxlib.protocols.ethernet.mdi.mdi1000base_t	MDI1000BaseT

For protocol-specific timing parameters (skew, loss, impedance per standard version), see references/protocol-standards.md.

Via Structures in Topologies

Via structures let the constraint engine track signal transitions between layers. They are Circuit subclasses with sig_in, sig_out, and COMMON ports.

from jitxlib.via_structures import SingleViaStructure, PolarViaGroundCage, SimpleAntiPad

These DO NOT EXIST — never import: jitx.via_structures, jitx.si.ViaStructure, jitxlib.vias

Simple via (bare signal transition)

from jitxlib.via_structures import SingleViaStructure

# Bare via — no ground cage or antipads
self.via = SingleViaStructure(
    MySubstrate.MicroVia,
    ground_cages=[],
    antipads=[],
    insertion_points=[],
)
self.GND += self.via.COMMON

# Chain topology through via
self += self.driver.out >> self.via.sig_in
self += self.via.sig_out >> self.receiver.inp

# Constrain full path (via is transparent to the constraint)
topo = Topology(self.driver.out, self.receiver.inp)
rs50 = current.substrate.routing_structure(50.0)
with ReferencePlanes(self.GND):
    self.cst = Constrain(topo).structure(rs50).insertion_loss(3.0)

RF via with ground cage

For controlled-impedance via transitions, add a PolarViaGroundCage and optional SimpleAntiPad:

from jitxlib.via_structures import SingleViaStructure, PolarViaGroundCage, SimpleAntiPad

self.rf_via = SingleViaStructure(
    MySubstrate.MicroviaL1L2,
    ground_cages=[
        PolarViaGroundCage(
            MySubstrate.BlindViaL1L4,  # Via type for cage
            count=12,                   # Number of ground vias
            radius=0.6,                 # Distance from signal via (mm)
        )
    ],
    antipads=[SimpleAntiPad(shape, layers)],
    insertion_points=[],
)
self.GND += self.rf_via.COMMON

# Same topology pattern as bare via
self += self.driver.out >> self.rf_via.sig_in
self += self.rf_via.sig_out >> self.receiver.inp

topo = Topology(self.driver.out, self.receiver.inp)
rs50 = current.substrate.routing_structure(50.0)
with ReferencePlanes(self.GND):
    self.cst = Constrain(topo).insertion_loss(1.0).structure(rs50)

Key points

• sig_in / sig_out — Signal enters and exits the via structure. Chain these with >> into your topology.
• COMMON — Ground connection for the cage vias. Always connect to your GND net with +=.
• insertion_points=[] — Required parameter. Pass empty list unless using custom insertion point geometry.
• Positioning — Via structures support .at(x, y, rotate=angle) for placement.
• DifferentialViaStructure — Same pattern but with DiffPair ports (sig_in.p, sig_in.n, etc.) and a pitch parameter for P/N spacing.

For RoutingStructure, DifferentialRoutingStructure, and substrate via definitions, see the jitx-substrate-modeler skill.

Common Mistakes

# WRONG: Using + for SI-constrained signals (no topology created)
self.data_net = self.src.data + self.dst.data

# CORRECT: Use >> for topology
self += self.src.data >> self.dst.data

# WRONG: Topology not stored (silently dropped)
self.src.data >> self.dst.data

# CORRECT: Store with += or named attribute
self += self.src.data >> self.dst.data

# WRONG: Using = for connections (Python assignment, no electrical connection)
self.src.data = self.dst.data

# WRONG: Constrain without ReferencePlanes when routing structure needs it
Constrain(topo).structure(rs50_with_reference)

# CORRECT: Wrap in ReferencePlanes
with ReferencePlanes(self.GND):
    Constrain(topo).structure(rs50_with_reference)

# WRONG: Creating Topology before creating the >> connection
topo = Topology(self.src, self.dst)  # Path doesn't exist yet!
self += self.src >> self.dst

# CORRECT: Create >> first, then identify with Topology
self += self.src >> self.dst
topo = Topology(self.src, self.dst)

Verification

Step 1: Type Check

pyright path/to/circuit.py

Step 2: Build Test

jitx build <module.path.DesignClass>

Don't run parallel JITX builds against the same project — sequence them. See jitx/SKILL.md "Build Safety".

SI constraint violations appear in the Issues List under "Unsatisfied Signal Constraints" in the JITX UI.

API Reference

For complete class definitions, all parameters, and method signatures:

• jitx.si module
• SI Topology concepts
• SI Constraints guide

Formatting

ruff format path/to/circuit.py