Spintronic Mapper¶

Maps SC networks onto magnetic tunnel junction (MTJ) arrays. Models spin-torque switching (STT/SOT), TMR calculation, and MuMax3 co-simulation.

Quick Start¶

Python

from sc_neurocore.spintronic.spintronic_mapper import (
    SpintronicMapper, MagneticDomainSim, SpinTorqueModel,
)

`sc_neurocore.spintronic.spintronic_mapper` ¶

Maps SC bitstreams to spintronic domain-wall and skyrmion logic.

Bridges deterministic SC arithmetic to magnetic-domain computing: domain-wall motion, skyrmion nucleation/annihilation, and spin-orbit torque (SOT) switching. Includes device variability models and co-simulation hooks for micromagnetic solvers (MuMax3).

Device technologies: - Domain-Wall (DW): Nanotrack racetrack with SOT-driven motion - Skyrmion (SKY): Skyrmion nucleation/annihilation gates - STT-MTJ: Spin-transfer torque magnetic tunnel junctions - SOT-MRAM: Spin-orbit torque MRAM cells

Pipeline

SC bitstream → SpintronicMapper → DeviceArray → MuMax3 co-sim

Compatible with: - memristor/memristor_mapper.py — shares variability injection pattern - hdl_gen/verilog_generator.py — RTL target for spintronic arrays - chiplet_gen/ — spintronic dies as chiplet nodes

`MaterialParams` `dataclass` ¶

Micromagnetic material parameters.

Source code in src/sc_neurocore/spintronic/spintronic_mapper.py

Python
@dataclass
class MaterialParams:
    """Micromagnetic material parameters."""

    saturation_magnetisation_a_m: float  # A/m (Ms)
    exchange_stiffness_j_m: float  # J/m (Aex)
    dmi_strength_j_m2: float  # J/m² (D, Dzyaloshinskii-Moriya)
    perpendicular_anisotropy_j_m3: float  # J/m³ (Ku)
    damping_alpha: float  # Gilbert damping
    temperature_k: float = 300.0

    @classmethod
    def cofeb_mgo(cls) -> MaterialParams:
        """CoFeB/MgO — standard STT-MTJ / SOT-MRAM stack."""
        return cls(
            saturation_magnetisation_a_m=1.2e6,
            exchange_stiffness_j_m=1.5e-11,
            dmi_strength_j_m2=0.0,
            perpendicular_anisotropy_j_m3=8e5,
            damping_alpha=0.01,
        )

    @classmethod
    def pt_co_multilayer(cls) -> MaterialParams:
        """Pt/Co multilayer — skyrmion host."""
        return cls(
            saturation_magnetisation_a_m=5.8e5,
            exchange_stiffness_j_m=1.5e-11,
            dmi_strength_j_m2=3.5e-3,
            perpendicular_anisotropy_j_m3=6e5,
            damping_alpha=0.015,
        )

    @classmethod
    def w_cofeb(cls) -> MaterialParams:
        """W/CoFeB — SOT switching with heavy-metal underlayer."""
        return cls(
            saturation_magnetisation_a_m=1.1e6,
            exchange_stiffness_j_m=1.3e-11,
            dmi_strength_j_m2=0.5e-3,
            perpendicular_anisotropy_j_m3=7e5,
            damping_alpha=0.02,
        )

`cofeb_mgo()` `classmethod` ¶

CoFeB/MgO — standard STT-MTJ / SOT-MRAM stack.

Source code in src/sc_neurocore/spintronic/spintronic_mapper.py

Python
@classmethod
def cofeb_mgo(cls) -> MaterialParams:
    """CoFeB/MgO — standard STT-MTJ / SOT-MRAM stack."""
    return cls(
        saturation_magnetisation_a_m=1.2e6,
        exchange_stiffness_j_m=1.5e-11,
        dmi_strength_j_m2=0.0,
        perpendicular_anisotropy_j_m3=8e5,
        damping_alpha=0.01,
    )

`pt_co_multilayer()` `classmethod` ¶

Pt/Co multilayer — skyrmion host.

Source code in src/sc_neurocore/spintronic/spintronic_mapper.py

Python
@classmethod
def pt_co_multilayer(cls) -> MaterialParams:
    """Pt/Co multilayer — skyrmion host."""
    return cls(
        saturation_magnetisation_a_m=5.8e5,
        exchange_stiffness_j_m=1.5e-11,
        dmi_strength_j_m2=3.5e-3,
        perpendicular_anisotropy_j_m3=6e5,
        damping_alpha=0.015,
    )

`w_cofeb()` `classmethod` ¶

W/CoFeB — SOT switching with heavy-metal underlayer.

Source code in src/sc_neurocore/spintronic/spintronic_mapper.py

Python
@classmethod
def w_cofeb(cls) -> MaterialParams:
    """W/CoFeB — SOT switching with heavy-metal underlayer."""
    return cls(
        saturation_magnetisation_a_m=1.1e6,
        exchange_stiffness_j_m=1.3e-11,
        dmi_strength_j_m2=0.5e-3,
        perpendicular_anisotropy_j_m3=7e5,
        damping_alpha=0.02,
    )

`SpintronicDeviceConfig` `dataclass` ¶

Configuration for a single spintronic device.

Source code in src/sc_neurocore/spintronic/spintronic_mapper.py

Python
@dataclass
class SpintronicDeviceConfig:
    """Configuration for a single spintronic device."""

    tech: SpintronicTech = SpintronicTech.SOT_MRAM
    material: MaterialParams = field(default_factory=MaterialParams.cofeb_mgo)
    width_nm: float = 80.0
    length_nm: float = 200.0
    thickness_nm: float = 1.2
    switching_current_ua: float = 50.0
    switching_time_ns: float = 1.0
    retention_years: float = 10.0
    tmr_ratio: float = 1.5
    error_rate: float = 1e-6

    @classmethod
    def from_tech(cls, tech: SpintronicTech) -> SpintronicDeviceConfig:
        presets = {
            SpintronicTech.DOMAIN_WALL: dict(
                material=MaterialParams.pt_co_multilayer(),
                width_nm=60.0,
                length_nm=1000.0,
                thickness_nm=0.8,
                switching_current_ua=100.0,
                switching_time_ns=5.0,
            ),
            SpintronicTech.SKYRMION: dict(
                material=MaterialParams.pt_co_multilayer(),
                width_nm=50.0,
                length_nm=500.0,
                thickness_nm=0.8,
                switching_current_ua=30.0,
                switching_time_ns=2.0,
            ),
            SpintronicTech.STT_MTJ: dict(
                material=MaterialParams.cofeb_mgo(),
                width_nm=40.0,
                length_nm=40.0,
                thickness_nm=1.2,
                switching_current_ua=80.0,
                switching_time_ns=3.0,
            ),
            SpintronicTech.SOT_MRAM: dict(
                material=MaterialParams.w_cofeb(),
                width_nm=80.0,
                length_nm=200.0,
                thickness_nm=1.0,
                switching_current_ua=50.0,
                switching_time_ns=0.5,
            ),
        }
        return cls(tech=tech, **presets[tech])

    @property
    def area_nm2(self) -> float:
        return self.width_nm * self.length_nm

    @property
    def switching_energy_fj(self) -> float:
        """E = I² × R × t (approximate, R ≈ 10 kΩ for MTJ)."""
        r_ohm = 10000.0
        i_a = self.switching_current_ua * 1e-6
        return i_a**2 * r_ohm * self.switching_time_ns * 1e6  # fJ

    @property
    def thermal_stability(self) -> float:
        """Thermal stability factor Δ = Ku × V / (kB × T).

        Needs to be > 40 for 10-year retention.
        """
        kb = 1.38064852e-23
        volume_m3 = (self.width_nm * self.length_nm * self.thickness_nm) * 1e-27
        t = self.material.temperature_k
        return self.material.perpendicular_anisotropy_j_m3 * volume_m3 / (kb * t)

    @property
    def read_disturb_probability(self) -> float:
        """Probability of read disturb (accidental flip during read).

        Approximation: P_rd ∝ exp(-Δ) for in-plane STT read.
        """
        delta = self.thermal_stability
        return float(np.exp(-delta)) if delta < 100 else 0.0

    @property
    def endurance_cycles(self) -> int:
        """Estimated write endurance cycles."""
        endurance_map = {
            SpintronicTech.DOMAIN_WALL: 10**15,
            SpintronicTech.SKYRMION: 10**15,
            SpintronicTech.STT_MTJ: 10**12,
            SpintronicTech.SOT_MRAM: 10**15,
        }
        return endurance_map.get(self.tech, 10**12)

`switching_energy_fj` `property` ¶

E = I² × R × t (approximate, R ≈ 10 kΩ for MTJ).

`thermal_stability` `property` ¶

Thermal stability factor Δ = Ku × V / (kB × T).

Needs to be > 40 for 10-year retention.

`read_disturb_probability` `property` ¶

Probability of read disturb (accidental flip during read).

Approximation: P_rd ∝ exp(-Δ) for in-plane STT read.

`endurance_cycles` `property` ¶

Estimated write endurance cycles.

`VariabilityModel` `dataclass` ¶

Process variability for spintronic devices.

Models dimension, anisotropy, and DMI variations from fab data.

Source code in src/sc_neurocore/spintronic/spintronic_mapper.py

Python
@dataclass
class VariabilityModel:
    """Process variability for spintronic devices.

    Models dimension, anisotropy, and DMI variations from fab data.
    """

    width_sigma_pct: float = 3.0
    length_sigma_pct: float = 3.0
    ku_sigma_pct: float = 5.0
    dmi_sigma_pct: float = 8.0
    damping_sigma_pct: float = 10.0
    ms_sigma_pct: float = 2.0

    def apply(
        self, device: SpintronicDeviceConfig, rng: np.random.Generator
    ) -> SpintronicDeviceConfig:
        """Return a variability-injected copy of the device."""
        import copy

        d = copy.deepcopy(device)
        d.width_nm *= 1 + rng.normal(0, self.width_sigma_pct / 100)
        d.length_nm *= 1 + rng.normal(0, self.length_sigma_pct / 100)
        d.material.perpendicular_anisotropy_j_m3 *= 1 + rng.normal(0, self.ku_sigma_pct / 100)
        d.material.dmi_strength_j_m2 *= 1 + rng.normal(0, self.dmi_sigma_pct / 100)
        d.material.damping_alpha *= 1 + rng.normal(0, self.damping_sigma_pct / 100)
        d.material.saturation_magnetisation_a_m *= 1 + rng.normal(0, self.ms_sigma_pct / 100)
        d.width_nm = max(10.0, d.width_nm)
        d.length_nm = max(10.0, d.length_nm)
        d.material.damping_alpha = max(0.001, d.material.damping_alpha)
        return d

`apply(device, rng)` ¶

Return a variability-injected copy of the device.

Source code in src/sc_neurocore/spintronic/spintronic_mapper.py

Python
def apply(
    self, device: SpintronicDeviceConfig, rng: np.random.Generator
) -> SpintronicDeviceConfig:
    """Return a variability-injected copy of the device."""
    import copy

    d = copy.deepcopy(device)
    d.width_nm *= 1 + rng.normal(0, self.width_sigma_pct / 100)
    d.length_nm *= 1 + rng.normal(0, self.length_sigma_pct / 100)
    d.material.perpendicular_anisotropy_j_m3 *= 1 + rng.normal(0, self.ku_sigma_pct / 100)
    d.material.dmi_strength_j_m2 *= 1 + rng.normal(0, self.dmi_sigma_pct / 100)
    d.material.damping_alpha *= 1 + rng.normal(0, self.damping_sigma_pct / 100)
    d.material.saturation_magnetisation_a_m *= 1 + rng.normal(0, self.ms_sigma_pct / 100)
    d.width_nm = max(10.0, d.width_nm)
    d.length_nm = max(10.0, d.length_nm)
    d.material.damping_alpha = max(0.001, d.material.damping_alpha)
    return d

`SpintronicCell` `dataclass` ¶

One cell in a spintronic array.

Source code in src/sc_neurocore/spintronic/spintronic_mapper.py

Python
@dataclass
class SpintronicCell:
    """One cell in a spintronic array."""

    row: int
    col: int
    device: SpintronicDeviceConfig
    state: int = 0  # 0 = P (parallel), 1 = AP (anti-parallel)
    weight_q88: int = 256  # Q8.8 = 1.0

    @property
    def resistance_ohm(self) -> float:
        r_p = 5000.0  # parallel resistance
        return r_p * (1 + self.state * self.device.tmr_ratio)

`SpintronicArray` ¶

Crossbar array of spintronic devices for SC computation.

Source code in src/sc_neurocore/spintronic/spintronic_mapper.py

Python
class SpintronicArray:
    """Crossbar array of spintronic devices for SC computation."""

    def __init__(
        self,
        rows: int,
        cols: int,
        tech: SpintronicTech = SpintronicTech.SOT_MRAM,
        variability: Optional[VariabilityModel] = None,
        rng_seed: int = 42,
    ):
        self.rows = rows
        self.cols = cols
        self.tech = tech
        self.rng = np.random.default_rng(rng_seed)
        base_device = SpintronicDeviceConfig.from_tech(tech)
        var = variability or VariabilityModel()
        self.cells: List[List[SpintronicCell]] = []
        for r in range(rows):
            row = []
            for c in range(cols):
                dev = var.apply(base_device, self.rng)
                row.append(SpintronicCell(r, c, dev))
            self.cells.append(row)

    @property
    def total_cells(self) -> int:
        return self.rows * self.cols

    @property
    def total_area_um2(self) -> float:
        return sum(c.device.area_nm2 for row in self.cells for c in row) / 1e6

    def program_weights(self, weights_q88: np.ndarray) -> None:
        """Program Q8.8 weights into the array."""
        for r in range(min(self.rows, weights_q88.shape[0])):
            for c in range(min(self.cols, weights_q88.shape[1])):
                w = int(weights_q88[r, c])
                self.cells[r][c].weight_q88 = w
                self.cells[r][c].state = 1 if w > 128 else 0

    def read_weights(self) -> np.ndarray:
        """Read weights back from the array."""
        w = np.zeros((self.rows, self.cols), dtype=np.int32)
        for r in range(self.rows):
            for c in range(self.cols):
                w[r, c] = self.cells[r][c].weight_q88
        return w

    def power_breakdown(self, bitstream_length: int = 256) -> Dict[str, float]:
        """Per-array power breakdown in femtojoules."""
        switch_energy = (
            sum(c.device.switching_energy_fj for row in self.cells for c in row) * bitstream_length
        )
        leakage_fj = (
            sum(
                1.0 / c.resistance_ohm * 0.1  # 100 mV read bias, 1 ns
                for row in self.cells
                for c in row
            )
            * bitstream_length
            * 1e6
        )
        return {
            "switching_fj": switch_energy,
            "leakage_fj": leakage_fj,
            "total_fj": switch_energy + leakage_fj,
        }

`program_weights(weights_q88)` ¶

Program Q8.8 weights into the array.

Source code in src/sc_neurocore/spintronic/spintronic_mapper.py

Python
def program_weights(self, weights_q88: np.ndarray) -> None:
    """Program Q8.8 weights into the array."""
    for r in range(min(self.rows, weights_q88.shape[0])):
        for c in range(min(self.cols, weights_q88.shape[1])):
            w = int(weights_q88[r, c])
            self.cells[r][c].weight_q88 = w
            self.cells[r][c].state = 1 if w > 128 else 0

`read_weights()` ¶

Read weights back from the array.

Source code in src/sc_neurocore/spintronic/spintronic_mapper.py

Python
def read_weights(self) -> np.ndarray:
    """Read weights back from the array."""
    w = np.zeros((self.rows, self.cols), dtype=np.int32)
    for r in range(self.rows):
        for c in range(self.cols):
            w[r, c] = self.cells[r][c].weight_q88
    return w

`power_breakdown(bitstream_length=256)` ¶

Per-array power breakdown in femtojoules.

Source code in src/sc_neurocore/spintronic/spintronic_mapper.py

Python
def power_breakdown(self, bitstream_length: int = 256) -> Dict[str, float]:
    """Per-array power breakdown in femtojoules."""
    switch_energy = (
        sum(c.device.switching_energy_fj for row in self.cells for c in row) * bitstream_length
    )
    leakage_fj = (
        sum(
            1.0 / c.resistance_ohm * 0.1  # 100 mV read bias, 1 ns
            for row in self.cells
            for c in row
        )
        * bitstream_length
        * 1e6
    )
    return {
        "switching_fj": switch_energy,
        "leakage_fj": leakage_fj,
        "total_fj": switch_energy + leakage_fj,
    }

`MappingResult` `dataclass` ¶

Result of mapping SC bitstreams to a spintronic array.

Source code in src/sc_neurocore/spintronic/spintronic_mapper.py

Python
@dataclass
class MappingResult:
    """Result of mapping SC bitstreams to a spintronic array."""

    array_rows: int
    array_cols: int
    tech: SpintronicTech
    total_area_um2: float
    total_energy_fj: float
    total_switching_time_ns: float
    bit_error_rate: float
    mc_yield: float = 1.0

`SpintronicMapper` ¶

Maps SC bitstreams + weights to a spintronic crossbar.

Performs: 1. Weight quantisation to device states (P/AP) 2. Array sizing from network dimensions 3. Energy/timing estimation 4. Monte-Carlo variability yield analysis

Source code in src/sc_neurocore/spintronic/spintronic_mapper.py

Python
class SpintronicMapper:
    """Maps SC bitstreams + weights to a spintronic crossbar.

    Performs:
    1. Weight quantisation to device states (P/AP)
    2. Array sizing from network dimensions
    3. Energy/timing estimation
    4. Monte-Carlo variability yield analysis
    """

    def __init__(
        self,
        tech: SpintronicTech = SpintronicTech.SOT_MRAM,
        variability: Optional[VariabilityModel] = None,
        rng_seed: int = 42,
    ):
        self.tech = tech
        self.variability = variability or VariabilityModel()
        self.rng = np.random.default_rng(rng_seed)

    def map_network(
        self,
        weights_q88: np.ndarray,
        bitstream_length: int = 256,
    ) -> Tuple[SpintronicArray, MappingResult]:
        """Map a weight matrix to a spintronic array."""
        rows, cols = weights_q88.shape
        array = SpintronicArray(
            rows,
            cols,
            self.tech,
            self.variability,
            self.rng.integers(0, 2**31),
        )
        array.program_weights(weights_q88)

        base = SpintronicDeviceConfig.from_tech(self.tech)
        total_e = base.switching_energy_fj * rows * cols * bitstream_length
        total_t = base.switching_time_ns * bitstream_length
        ber = base.error_rate * rows * cols

        return array, MappingResult(
            rows,
            cols,
            self.tech,
            array.total_area_um2,
            total_e,
            total_t,
            ber,
        )

    def monte_carlo_yield(
        self,
        weights_q88: np.ndarray,
        n_trials: int = 100,
        tolerance_q88: int = 16,
    ) -> float:
        """Run Monte-Carlo yield analysis."""
        passing = 0
        for _ in range(n_trials):
            seed = int(self.rng.integers(0, 2**31))
            array = SpintronicArray(
                weights_q88.shape[0],
                weights_q88.shape[1],
                self.tech,
                self.variability,
                seed,
            )
            array.program_weights(weights_q88)
            readback = array.read_weights()
            max_error = int(
                np.max(np.abs(readback.astype(np.int32) - weights_q88.astype(np.int32)))
            )
            if max_error <= tolerance_q88:
                passing += 1
        return passing / n_trials

`map_network(weights_q88, bitstream_length=256)` ¶

Map a weight matrix to a spintronic array.

Source code in src/sc_neurocore/spintronic/spintronic_mapper.py

Python
def map_network(
    self,
    weights_q88: np.ndarray,
    bitstream_length: int = 256,
) -> Tuple[SpintronicArray, MappingResult]:
    """Map a weight matrix to a spintronic array."""
    rows, cols = weights_q88.shape
    array = SpintronicArray(
        rows,
        cols,
        self.tech,
        self.variability,
        self.rng.integers(0, 2**31),
    )
    array.program_weights(weights_q88)

    base = SpintronicDeviceConfig.from_tech(self.tech)
    total_e = base.switching_energy_fj * rows * cols * bitstream_length
    total_t = base.switching_time_ns * bitstream_length
    ber = base.error_rate * rows * cols

    return array, MappingResult(
        rows,
        cols,
        self.tech,
        array.total_area_um2,
        total_e,
        total_t,
        ber,
    )

`monte_carlo_yield(weights_q88, n_trials=100, tolerance_q88=16)` ¶

Run Monte-Carlo yield analysis.

Source code in src/sc_neurocore/spintronic/spintronic_mapper.py

Python
def monte_carlo_yield(
    self,
    weights_q88: np.ndarray,
    n_trials: int = 100,
    tolerance_q88: int = 16,
) -> float:
    """Run Monte-Carlo yield analysis."""
    passing = 0
    for _ in range(n_trials):
        seed = int(self.rng.integers(0, 2**31))
        array = SpintronicArray(
            weights_q88.shape[0],
            weights_q88.shape[1],
            self.tech,
            self.variability,
            seed,
        )
        array.program_weights(weights_q88)
        readback = array.read_weights()
        max_error = int(
            np.max(np.abs(readback.astype(np.int32) - weights_q88.astype(np.int32)))
        )
        if max_error <= tolerance_q88:
            passing += 1
    return passing / n_trials

`MuMax3ScriptGenerator` ¶

Generates MuMax3 (.mx3) scripts for micromagnetic co-simulation.

Source code in src/sc_neurocore/spintronic/spintronic_mapper.py

Python
class MuMax3ScriptGenerator:
    """Generates MuMax3 (.mx3) scripts for micromagnetic co-simulation."""

    @staticmethod
    def generate_switching(
        device: SpintronicDeviceConfig,
        current_density_a_m2: float = 1e12,
        duration_ns: float = 5.0,
    ) -> str:
        m = device.material
        return f"""\
// SC-NeuroCore MuMax3 Co-Simulation — SOT Switching
// Tech: {device.tech.value}

SetGridSize(128, 32, 1)
SetCellSize({device.width_nm / 128:.2f}e-9, {device.length_nm / 32:.2f}e-9, {device.thickness_nm:.1f}e-9)

Msat  = {m.saturation_magnetisation_a_m:.1f}
Aex   = {m.exchange_stiffness_j_m:.2e}
Ku1   = {m.perpendicular_anisotropy_j_m3:.2e}
AnisU = vector(0, 0, 1)
Dind  = {m.dmi_strength_j_m2:.2e}
alpha = {m.damping_alpha:.4f}

Temp = {m.temperature_k:.0f}

m = uniform(0, 0, 1)

J = vector({current_density_a_m2:.2e}, 0, 0)
xi = 0.05
Pol = 0.56

Run({duration_ns:.1f}e-9)

SaveAs(m, "final_state")
"""

    @staticmethod
    def generate_skyrmion(
        device: SpintronicDeviceConfig,
    ) -> str:
        m = device.material
        return f"""\
// SC-NeuroCore MuMax3 — Skyrmion Nucleation
SetGridSize(256, 256, 1)
SetCellSize({device.width_nm / 256:.2f}e-9, {device.width_nm / 256:.2f}e-9, {device.thickness_nm:.1f}e-9)

Msat  = {m.saturation_magnetisation_a_m:.1f}
Aex   = {m.exchange_stiffness_j_m:.2e}
Ku1   = {m.perpendicular_anisotropy_j_m3:.2e}
AnisU = vector(0, 0, 1)
Dind  = {m.dmi_strength_j_m2:.2e}
alpha = {m.damping_alpha:.4f}

Temp = {m.temperature_k:.0f}

m = uniform(0, 0, 1)
m.SetRegion(0, NeelSkyrmion(1, -1))

Relax()
SaveAs(m, "skyrmion_ground_state")
"""

`SpintronicVerilogGenerator` ¶

Generates Verilog wrappers for spintronic crossbar arrays.

Source code in src/sc_neurocore/spintronic/spintronic_mapper.py

Python
class SpintronicVerilogGenerator:
    """Generates Verilog wrappers for spintronic crossbar arrays."""

    @staticmethod
    def generate(
        array_name: str,
        rows: int,
        cols: int,
        tech: SpintronicTech,
    ) -> str:
        return f"""\
// SC-NeuroCore — Spintronic Array Wrapper ({tech.value})
// Auto-generated — do not edit

module {array_name} #(
    parameter ROWS = {rows},
    parameter COLS = {cols},
    parameter BITSTREAM_W = 256
)(
    input  wire                     clk,
    input  wire                     rst_n,
    input  wire [BITSTREAM_W-1:0]   bitstream_in  [0:COLS-1],
    output wire [BITSTREAM_W-1:0]   bitstream_out [0:ROWS-1],
    // Weight programming interface
    input  wire [$clog2(ROWS)-1:0]  prog_row,
    input  wire [$clog2(COLS)-1:0]  prog_col,
    input  wire [15:0]              prog_weight,  // Q8.8
    input  wire                     prog_en
);

    // Weight storage (maps to spintronic state via external driver)
    reg [15:0] weights [0:ROWS-1][0:COLS-1];

    // Programming
    always @(posedge clk or negedge rst_n) begin
        if (!rst_n) begin
            // No reset for NVM — retain state
        end else if (prog_en) begin
            weights[prog_row][prog_col] <= prog_weight;
        end
    end

    // SC dot product per output row
    genvar r, c;
    generate
        for (r = 0; r < ROWS; r = r + 1) begin : row_gen
            wire [BITSTREAM_W-1:0] partial [0:COLS-1];
            for (c = 0; c < COLS; c = c + 1) begin : col_gen
                // SC AND gate: bitstream × weight-encoded bitstream
                assign partial[c] = bitstream_in[c]; // Simplified; actual uses LFSR threshold
            end
            // OR-reduction (majority) of partial products
            assign bitstream_out[r] = partial[0]; // Placeholder for full OR tree
        end
    endgenerate

endmodule
"""

`RacetrackShiftRegister` `dataclass` ¶

Domain-wall racetrack memory with current-driven bit shifting.

Models a nanotrack with N bit positions; SOT current pulses shift all domain walls one position per clock edge.

Source code in src/sc_neurocore/spintronic/spintronic_mapper.py

Python
@dataclass
class RacetrackShiftRegister:
    """Domain-wall racetrack memory with current-driven bit shifting.

    Models a nanotrack with N bit positions; SOT current pulses shift
    all domain walls one position per clock edge.
    """

    n_positions: int
    bits: Optional[np.ndarray] = None
    shift_current_ua: float = 100.0
    shift_time_ns: float = 0.5
    shift_error_rate: float = 1e-5

    def __post_init__(self):
        if self.bits is None:
            self.bits = np.zeros(self.n_positions, dtype=np.int8)

    def load(self, data: np.ndarray) -> None:
        self.bits = np.array(data[: self.n_positions], dtype=np.int8)

    def shift_right(self, n: int = 1, rng: Optional[np.random.Generator] = None) -> None:
        for _ in range(n):
            self.bits = np.roll(self.bits, 1)
            self.bits[0] = 0
            if rng is not None and rng.random() < self.shift_error_rate:
                pos = rng.integers(0, self.n_positions)
                self.bits[pos] ^= 1

    def shift_left(self, n: int = 1, rng: Optional[np.random.Generator] = None) -> None:
        for _ in range(n):
            self.bits = np.roll(self.bits, -1)
            self.bits[-1] = 0
            if rng is not None and rng.random() < self.shift_error_rate:
                pos = rng.integers(0, self.n_positions)
                self.bits[pos] ^= 1

    @property
    def shift_energy_fj(self) -> float:
        r_ohm = 500.0
        i_a = self.shift_current_ua * 1e-6
        return i_a**2 * r_ohm * self.shift_time_ns * 1e6

`SkyrmionHallCorrector` `dataclass` ¶

Corrects skyrmion trajectory for the skyrmion Hall effect.

Skyrmions drift at an angle θ_H to the applied current direction. θ_H = arctan(4π Q / (α D)) where Q is topological charge.

Source code in src/sc_neurocore/spintronic/spintronic_mapper.py

Python
@dataclass
class SkyrmionHallCorrector:
    """Corrects skyrmion trajectory for the skyrmion Hall effect.

    Skyrmions drift at an angle θ_H to the applied current direction.
    θ_H = arctan(4π Q / (α D)) where Q is topological charge.
    """

    topological_charge: int = 1
    damping_alpha: float = 0.015
    dmi_strength: float = 3.5e-3

    @property
    def hall_angle_deg(self) -> float:
        ratio = 4 * math.pi * abs(self.topological_charge) * self.damping_alpha
        return math.degrees(math.atan(ratio))

    def corrected_position(self, x_drive: float, track_width_nm: float) -> Tuple[float, float]:
        """Return (x, y) position accounting for Hall drift."""
        theta = math.radians(self.hall_angle_deg)
        y_drift = x_drive * math.tan(theta)
        y_clamped = max(-track_width_nm / 2, min(track_width_nm / 2, y_drift))
        return (x_drive, y_clamped)

    @property
    def needs_confinement(self) -> bool:
        return self.hall_angle_deg > 5.0

`corrected_position(x_drive, track_width_nm)` ¶

Return (x, y) position accounting for Hall drift.

Source code in src/sc_neurocore/spintronic/spintronic_mapper.py

Python
def corrected_position(self, x_drive: float, track_width_nm: float) -> Tuple[float, float]:
    """Return (x, y) position accounting for Hall drift."""
    theta = math.radians(self.hall_angle_deg)
    y_drift = x_drive * math.tan(theta)
    y_clamped = max(-track_width_nm / 2, min(track_width_nm / 2, y_drift))
    return (x_drive, y_clamped)

`MLCConfig` `dataclass` ¶

Multi-level cell configuration for spintronic devices.

Source code in src/sc_neurocore/spintronic/spintronic_mapper.py

Python
@dataclass
class MLCConfig:
    """Multi-level cell configuration for spintronic devices."""

    bits_per_cell: int = 2
    levels: int = 0

    def __post_init__(self):
        self.levels = 2**self.bits_per_cell

    @property
    def resistance_margins(self) -> List[float]:
        """Resistance levels evenly spaced between R_P and R_AP."""
        r_p, r_ap = 5000.0, 12500.0
        step = (r_ap - r_p) / (self.levels - 1) if self.levels > 1 else 0
        return [r_p + i * step for i in range(self.levels)]

    def quantize_weight(self, weight_float: float) -> int:
        """Quantize a [0, 1] weight to an MLC level."""
        level = int(round(weight_float * (self.levels - 1)))
        return max(0, min(self.levels - 1, level))

    def dequantize(self, level: int) -> float:
        return level / (self.levels - 1) if self.levels > 1 else 0.0

    @property
    def density_improvement(self) -> float:
        return float(self.bits_per_cell)

`resistance_margins` `property` ¶

Resistance levels evenly spaced between R_P and R_AP.

`quantize_weight(weight_float)` ¶

Quantize a [0, 1] weight to an MLC level.

Source code in src/sc_neurocore/spintronic/spintronic_mapper.py

Python
def quantize_weight(self, weight_float: float) -> int:
    """Quantize a [0, 1] weight to an MLC level."""
    level = int(round(weight_float * (self.levels - 1)))
    return max(0, min(self.levels - 1, level))

`WriteVerifyResult` `dataclass` ¶

Result of a write-verify cycle.

Source code in src/sc_neurocore/spintronic/spintronic_mapper.py

Python
@dataclass
class WriteVerifyResult:
    """Result of a write-verify cycle."""

    target_weight: int
    actual_weight: int
    attempts: int
    success: bool

    @property
    def error(self) -> int:
        return abs(self.target_weight - self.actual_weight)

`AgingModel` `dataclass` ¶

Endurance degradation model for spintronic devices.

TMR and thermal stability degrade with write cycles.

Source code in src/sc_neurocore/spintronic/spintronic_mapper.py

Python
@dataclass
class AgingModel:
    """Endurance degradation model for spintronic devices.

    TMR and thermal stability degrade with write cycles.
    """

    cycles_written: int = 0

    def tmr_degradation(self, initial_tmr: float, endurance_limit: int) -> float:
        """TMR degrades linearly as cycling approaches endurance limit."""
        if endurance_limit <= 0:
            return initial_tmr
        frac = min(1.0, self.cycles_written / endurance_limit)
        return initial_tmr * (1.0 - 0.3 * frac)  # up to 30% TMR loss

    def stability_degradation(self, initial_delta: float, endurance_limit: int) -> float:
        """Thermal stability degrades with oxide breakdown."""
        if endurance_limit <= 0:
            return initial_delta
        frac = min(1.0, self.cycles_written / endurance_limit)
        return initial_delta * (1.0 - 0.2 * frac)

    @property
    def is_worn_out(self) -> bool:
        return self.cycles_written > 0 and self.tmr_degradation(1.5, 10**12) < 0.5

    def write(self, n: int = 1) -> None:
        self.cycles_written += n

`tmr_degradation(initial_tmr, endurance_limit)` ¶

TMR degrades linearly as cycling approaches endurance limit.

Source code in src/sc_neurocore/spintronic/spintronic_mapper.py

Python
def tmr_degradation(self, initial_tmr: float, endurance_limit: int) -> float:
    """TMR degrades linearly as cycling approaches endurance limit."""
    if endurance_limit <= 0:
        return initial_tmr
    frac = min(1.0, self.cycles_written / endurance_limit)
    return initial_tmr * (1.0 - 0.3 * frac)  # up to 30% TMR loss

`stability_degradation(initial_delta, endurance_limit)` ¶

Thermal stability degrades with oxide breakdown.

Source code in src/sc_neurocore/spintronic/spintronic_mapper.py

Python
def stability_degradation(self, initial_delta: float, endurance_limit: int) -> float:
    """Thermal stability degrades with oxide breakdown."""
    if endurance_limit <= 0:
        return initial_delta
    frac = min(1.0, self.cycles_written / endurance_limit)
    return initial_delta * (1.0 - 0.2 * frac)

`RadiationModel` `dataclass` ¶

SEU and TID models for spintronic devices in radiation environments.

Source code in src/sc_neurocore/spintronic/spintronic_mapper.py

Python
@dataclass
class RadiationModel:
    """SEU and TID models for spintronic devices in radiation environments."""

    seu_cross_section_cm2: float = 1e-14  # per device
    tid_threshold_krad: float = 1000.0

    def seu_rate(self, flux_particles_cm2_s: float, n_devices: int) -> float:
        """SEU events per second."""
        return self.seu_cross_section_cm2 * flux_particles_cm2_s * n_devices

    def tid_degradation(self, dose_krad: float) -> float:
        """Fraction of performance remaining after TID."""
        if dose_krad >= self.tid_threshold_krad:
            return 0.5  # 50% degradation at threshold
        return 1.0 - 0.5 * (dose_krad / self.tid_threshold_krad)

    @property
    def is_rad_hard(self) -> bool:
        """Spintronic devices are inherently radiation-hard (non-charge-based)."""
        return self.tid_threshold_krad >= 100.0

`is_rad_hard` `property` ¶

Spintronic devices are inherently radiation-hard (non-charge-based).

`seu_rate(flux_particles_cm2_s, n_devices)` ¶

SEU events per second.

Source code in src/sc_neurocore/spintronic/spintronic_mapper.py

Python
def seu_rate(self, flux_particles_cm2_s: float, n_devices: int) -> float:
    """SEU events per second."""
    return self.seu_cross_section_cm2 * flux_particles_cm2_s * n_devices

`tid_degradation(dose_krad)` ¶

Fraction of performance remaining after TID.

Source code in src/sc_neurocore/spintronic/spintronic_mapper.py

Python
def tid_degradation(self, dose_krad: float) -> float:
    """Fraction of performance remaining after TID."""
    if dose_krad >= self.tid_threshold_krad:
        return 0.5  # 50% degradation at threshold
    return 1.0 - 0.5 * (dose_krad / self.tid_threshold_krad)

`DefectEntry` `dataclass` ¶

One defective cell in the array.

Source code in src/sc_neurocore/spintronic/spintronic_mapper.py

Python
@dataclass
class DefectEntry:
    """One defective cell in the array."""

    row: int
    col: int
    defect_type: str  # "stuck_p", "stuck_ap", "open", "short"

`DefectMap` ¶

Tracks and remaps defective cells in a spintronic array.

Source code in src/sc_neurocore/spintronic/spintronic_mapper.py

Python
class DefectMap:
    """Tracks and remaps defective cells in a spintronic array."""

    def __init__(self):
        self.defects: List[DefectEntry] = []
        self.remap: Dict[Tuple[int, int], Tuple[int, int]] = {}

    def add_defect(self, row: int, col: int, defect_type: str) -> None:
        self.defects.append(DefectEntry(row, col, defect_type))

    @property
    def defect_count(self) -> int:
        return len(self.defects)

    def defect_rate(self, total_cells: int) -> float:
        if total_cells <= 0:
            return 0.0
        return self.defect_count / total_cells

    def add_remap(self, bad: Tuple[int, int], spare: Tuple[int, int]) -> None:
        self.remap[bad] = spare

    def is_defective(self, row: int, col: int) -> bool:
        return any(d.row == row and d.col == col for d in self.defects)

    def effective_address(self, row: int, col: int) -> Tuple[int, int]:
        return self.remap.get((row, col), (row, col))

`MuMax3Result` `dataclass` ¶

Parsed result from a MuMax3 simulation.

Source code in src/sc_neurocore/spintronic/spintronic_mapper.py

Python
@dataclass
class MuMax3Result:
    """Parsed result from a MuMax3 simulation."""

    final_mx: float = 0.0
    final_my: float = 0.0
    final_mz: float = 0.0
    switched: bool = False
    energy_j: float = 0.0
    sim_time_ns: float = 0.0

    @property
    def magnetisation_magnitude(self) -> float:
        return math.sqrt(self.final_mx**2 + self.final_my**2 + self.final_mz**2)

`MuMax3OutputParser` ¶

Parses MuMax3 table output for co-simulation integration.

Source code in src/sc_neurocore/spintronic/spintronic_mapper.py

Python
class MuMax3OutputParser:
    """Parses MuMax3 table output for co-simulation integration."""

    @staticmethod
    def parse_table(text: str) -> MuMax3Result:
        """Parse a MuMax3 .table output (TSV with header)."""
        lines = [l.strip() for l in text.strip().split("\n") if l.strip() and not l.startswith("#")]
        if not lines:
            return MuMax3Result()
        last = lines[-1].split("\t")
        if len(last) < 4:
            last = lines[-1].split()
        try:
            t = float(last[0])
            mx = float(last[1])
            my = float(last[2])
            mz = float(last[3])
            switched = mz < 0  # switched if mz flipped
            return MuMax3Result(mx, my, mz, switched, sim_time_ns=t * 1e9)
        except (ValueError, IndexError):
            return MuMax3Result()

    @staticmethod
    def is_switching_successful(result: MuMax3Result) -> bool:
        return result.switched and result.magnetisation_magnitude > 0.9

`parse_table(text)` `staticmethod` ¶

Parse a MuMax3 .table output (TSV with header).

Source code in src/sc_neurocore/spintronic/spintronic_mapper.py

Python
@staticmethod
def parse_table(text: str) -> MuMax3Result:
    """Parse a MuMax3 .table output (TSV with header)."""
    lines = [l.strip() for l in text.strip().split("\n") if l.strip() and not l.startswith("#")]
    if not lines:
        return MuMax3Result()
    last = lines[-1].split("\t")
    if len(last) < 4:
        last = lines[-1].split()
    try:
        t = float(last[0])
        mx = float(last[1])
        my = float(last[2])
        mz = float(last[3])
        switched = mz < 0  # switched if mz flipped
        return MuMax3Result(mx, my, mz, switched, sim_time_ns=t * 1e9)
    except (ValueError, IndexError):
        return MuMax3Result()

`switching_current_vs_temperature(i_c0_ua, delta_0, temperature_k, temp_ref_k=300.0)` ¶

Ic(T) = Ic0 × (1 - T/T_ref × kBT/(Δ0 × kBT_ref)).

Simplified Néel-Arrhenius model for thermally activated switching.

Source code in src/sc_neurocore/spintronic/spintronic_mapper.py

Python
def switching_current_vs_temperature(
    i_c0_ua: float,
    delta_0: float,
    temperature_k: float,
    temp_ref_k: float = 300.0,
) -> float:
    """Ic(T) = Ic0 × (1 - T/T_ref × kBT/(Δ0 × kBT_ref)).

    Simplified Néel-Arrhenius model for thermally activated switching.
    """
    if temp_ref_k <= 0 or delta_0 <= 0:
        return i_c0_ua
    ratio = temperature_k / temp_ref_k
    factor = max(0.01, 1.0 - ratio * (1.0 / delta_0))
    return i_c0_ua * factor

`switching_time_vs_temperature(t_sw0_ns, temperature_k, temp_ref_k=300.0)` ¶

Switching time increases with temperature due to thermal fluctuations.

Source code in src/sc_neurocore/spintronic/spintronic_mapper.py

Python
def switching_time_vs_temperature(
    t_sw0_ns: float,
    temperature_k: float,
    temp_ref_k: float = 300.0,
) -> float:
    """Switching time increases with temperature due to thermal fluctuations."""
    ratio = temperature_k / temp_ref_k
    return t_sw0_ns * (1.0 + 0.1 * (ratio - 1.0))

`retention_failure_probability(thermal_stability, time_seconds, attempt_freq_hz=1000000000.0)` ¶

P_fail = 1 - exp(-t × f0 × exp(-Δ)).

Source code in src/sc_neurocore/spintronic/spintronic_mapper.py

Python
def retention_failure_probability(
    thermal_stability: float,
    time_seconds: float,
    attempt_freq_hz: float = 1e9,
) -> float:
    """P_fail = 1 - exp(-t × f0 × exp(-Δ))."""
    if thermal_stability > 100:
        return 0.0
    exponent = -thermal_stability
    rate = attempt_freq_hz * math.exp(exponent)
    p = 1.0 - math.exp(-time_seconds * rate)
    return max(0.0, min(1.0, p))

`write_verify(cell, target_q88, max_attempts=5, rng=None)` ¶

Program a cell with write-verify loop.

Source code in src/sc_neurocore/spintronic/spintronic_mapper.py

Python
def write_verify(
    cell: SpintronicCell,
    target_q88: int,
    max_attempts: int = 5,
    rng: Optional[np.random.Generator] = None,
) -> WriteVerifyResult:
    """Program a cell with write-verify loop."""
    for attempt in range(1, max_attempts + 1):
        cell.weight_q88 = target_q88
        cell.state = 1 if target_q88 > 128 else 0
        if rng is not None:
            noise = int(rng.normal(0, 2))
            cell.weight_q88 = max(0, min(511, cell.weight_q88 + noise))
        if abs(cell.weight_q88 - target_q88) <= 4:
            return WriteVerifyResult(target_q88, cell.weight_q88, attempt, True)
    return WriteVerifyResult(target_q88, cell.weight_q88, max_attempts, False)

Spintronic Mapper¶

Quick Start¶

sc_neurocore.spintronic.spintronic_mapper ¶

MaterialParams dataclass ¶

cofeb_mgo() classmethod ¶

pt_co_multilayer() classmethod ¶

w_cofeb() classmethod ¶

SpintronicDeviceConfig dataclass ¶

switching_energy_fj property ¶

thermal_stability property ¶

read_disturb_probability property ¶

endurance_cycles property ¶

VariabilityModel dataclass ¶

apply(device, rng) ¶

SpintronicCell dataclass ¶

SpintronicArray ¶

program_weights(weights_q88) ¶

read_weights() ¶

power_breakdown(bitstream_length=256) ¶

MappingResult dataclass ¶

SpintronicMapper ¶

map_network(weights_q88, bitstream_length=256) ¶

monte_carlo_yield(weights_q88, n_trials=100, tolerance_q88=16) ¶

MuMax3ScriptGenerator ¶

SpintronicVerilogGenerator ¶

RacetrackShiftRegister dataclass ¶

SkyrmionHallCorrector dataclass ¶

corrected_position(x_drive, track_width_nm) ¶

MLCConfig dataclass ¶

resistance_margins property ¶

quantize_weight(weight_float) ¶

WriteVerifyResult dataclass ¶

AgingModel dataclass ¶

tmr_degradation(initial_tmr, endurance_limit) ¶

stability_degradation(initial_delta, endurance_limit) ¶

RadiationModel dataclass ¶

is_rad_hard property ¶

seu_rate(flux_particles_cm2_s, n_devices) ¶

tid_degradation(dose_krad) ¶

DefectEntry dataclass ¶

DefectMap ¶

MuMax3Result dataclass ¶

MuMax3OutputParser ¶

parse_table(text) staticmethod ¶

switching_current_vs_temperature(i_c0_ua, delta_0, temperature_k, temp_ref_k=300.0) ¶

switching_time_vs_temperature(t_sw0_ns, temperature_k, temp_ref_k=300.0) ¶

retention_failure_probability(thermal_stability, time_seconds, attempt_freq_hz=1000000000.0) ¶

write_verify(cell, target_q88, max_attempts=5, rng=None) ¶

`sc_neurocore.spintronic.spintronic_mapper` ¶

`MaterialParams` `dataclass` ¶

`cofeb_mgo()` `classmethod` ¶

`pt_co_multilayer()` `classmethod` ¶

`w_cofeb()` `classmethod` ¶

`SpintronicDeviceConfig` `dataclass` ¶

`switching_energy_fj` `property` ¶

`thermal_stability` `property` ¶

`read_disturb_probability` `property` ¶

`endurance_cycles` `property` ¶

`VariabilityModel` `dataclass` ¶

`apply(device, rng)` ¶

`SpintronicCell` `dataclass` ¶

`SpintronicArray` ¶

`program_weights(weights_q88)` ¶

`read_weights()` ¶

`power_breakdown(bitstream_length=256)` ¶

`MappingResult` `dataclass` ¶

`SpintronicMapper` ¶

`map_network(weights_q88, bitstream_length=256)` ¶

`monte_carlo_yield(weights_q88, n_trials=100, tolerance_q88=16)` ¶

`MuMax3ScriptGenerator` ¶

`SpintronicVerilogGenerator` ¶

`RacetrackShiftRegister` `dataclass` ¶

`SkyrmionHallCorrector` `dataclass` ¶

`corrected_position(x_drive, track_width_nm)` ¶

`MLCConfig` `dataclass` ¶

`resistance_margins` `property` ¶

`quantize_weight(weight_float)` ¶

`WriteVerifyResult` `dataclass` ¶

`AgingModel` `dataclass` ¶

`tmr_degradation(initial_tmr, endurance_limit)` ¶

`stability_degradation(initial_delta, endurance_limit)` ¶

`RadiationModel` `dataclass` ¶

`is_rad_hard` `property` ¶

`seu_rate(flux_particles_cm2_s, n_devices)` ¶

`tid_degradation(dose_krad)` ¶

`DefectEntry` `dataclass` ¶

`DefectMap` ¶

`MuMax3Result` `dataclass` ¶

`MuMax3OutputParser` ¶

`parse_table(text)` `staticmethod` ¶

`switching_current_vs_temperature(i_c0_ua, delta_0, temperature_k, temp_ref_k=300.0)` ¶

`switching_time_vs_temperature(t_sw0_ns, temperature_k, temp_ref_k=300.0)` ¶

`retention_failure_probability(thermal_stability, time_seconds, attempt_freq_hz=1000000000.0)` ¶

`write_verify(cell, target_q88, max_attempts=5, rng=None)` ¶