Skip to content

Research Hardware Platforms Guide

SC-NeuroCore currently registers 194 hardware profiles across 38 platform classes. The registry spans stable deployment targets, research accelerators, and explicitly experimental substrate profiles, from FPGAs and ASICs to organoid, molecular, and quantum-neuromorphic research targets.

This guide covers the research platform classes, their physical principles, compilation considerations, and code examples.

Table of Contents

  1. Cryogenic and Non-Volatile Platforms
  2. Biological, Electrochemical, and Wafer-Scale Platforms
  3. Memory-Centric and Quantum-Inspired Platforms
  4. Interconnect, Acoustic, Fluidic, and Space Platforms
  5. Sovereign, Organic, and Magnonic Platforms
  6. Adaptive Reliability Platforms
  7. Custom Profile Registration
  8. Platform Class Reference

Cryogenic and Non-Volatile Platforms

Superconducting / Cryogenic (3 profiles)

Physics: Single Flux Quantum (SFQ) logic operates at 4K with Josephson junctions producing picosecond voltage pulses. Clock speeds reach 100+ GHz with near-zero switching energy (~10⁻¹⁹ J per gate).

Profile Vendor Family Width
nist_sfq NIST SFQ 8-bit
northrop_aqfp Northrop Grumman AQFP 8-bit
mit_josephson MIT-LL Josephson-JJ 8-bit

Compilation notes: - Use 8-bit or narrower data widths (gate budgets are small) - Overflow: saturate (no wrap — metastability risk at 4K) - SFQ pulse timing requires careful pipeline depth

Python
from sc_neurocore.compiler.platforms import get_profile

p = get_profile("nist_sfq")
print(f"Class: {p.platform_class}")  # superconducting
print(f"Width: Q{p.data_width - p.fraction}.{p.fraction}")  # Q4.4

Spintronic (2 profiles)

Physics: Spin-Transfer Torque (STT) and Spin-Orbit Torque (SOT) MRAM use electron spin states for non-volatile storage. Enables in-memory computation with ns switching and near-infinite endurance.

Profile Vendor Family Width
everspin_stt Everspin STT-MRAM 8-bit
samsung_sot Samsung SOT-MRAM 8-bit

Key advantage: Non-volatile synaptic weights survive power cycles, enabling instant-on neural inference without weight reload.

Ferroelectric (2 profiles)

Physics: Ferroelectric FETs (FeFET) and FeRAM use polarisation switching in hafnium oxide (HfO₂) for non-volatile CIM with CMOS-compatible process integration.

Profile Vendor Family Width
gf_fefet GlobalFoundries FeFET-22FDX 8-bit
sk_hynix_feram SK Hynix FeRAM 8-bit

CGRA (3 profiles)

Physics: Coarse-Grained Reconfigurable Arrays provide FPGA-like flexibility with ASIC-like efficiency. Reconfiguration happens at the PE (Processing Element) level, not the LUT level.

Profile Vendor Family Width Freq
samsung_cgra Samsung CGRA-v2 16-bit 1 GHz
qualcomm_cgra Qualcomm NPU-CGRA 8-bit 2 GHz
cadence_xtensa Cadence Xtensa-NX 16-bit 1.5 GHz

3D-Stacked (3 profiles)

Physics: 3D integration via through-silicon vias (TSVs), hybrid bonding, or micro-bumps. Dramatically reduces interconnect length and enables heterogeneous integration (logic + memory on same stack).

Profile Vendor Family Width Freq
tsmc_soic TSMC SoIC-3D 16-bit 2 GHz
intel_foveros Intel Foveros-3D 16-bit 3 GHz
amd_3d_vcache AMD 3D-V-Cache 16-bit 4.5 GHz

Edge MCU / TinyML (5 profiles)

Physics: Standard ARM Cortex-M and RISC-V microcontrollers with hardware neural accelerators. Target: <1mW inference at the sensor edge.

Profile Vendor Family Width Freq
rp2040 Raspberry Pi RP2040 16-bit 133 MHz
esp32_s3 Espressif ESP32-S3 16-bit 240 MHz
stm32_h7 STMicro STM32H7 16-bit 480 MHz
nrf5340 Nordic nRF5340 16-bit 128 MHz
max78000 Maxim/ADI MAX78000 8-bit 100 MHz
Python
from sc_neurocore.compiler.platforms import get_profile

p = get_profile("max78000")
print(f"Built-in CNN accelerator: {p.notes}")

Biological, Electrochemical, and Wafer-Scale Platforms

Biological / Wetware (2 profiles)

Physics: Living organoid co-processors. FinalSpark uses human iPSC neuron cultures as computational substrates. Cortical Labs (DishBrain) demonstrated game-playing biological neural networks.

Profile Vendor Family Width
finalspark_neuroplatform FinalSpark Neuroplatform 16-bit
cortical_labs_dishbrain Cortical Labs DishBrain 16-bit

Compilation notes: - Output is stimulation protocol, not RTL - Time constants are biological (ms–s scale) - Stochastic behaviour — no deterministic guarantee

Electrochemical / Memristive (3 profiles)

Physics: Electrochemical RAM (ECRAM) and Phase-Change RAM (PCRAM) enable analog synaptic weight storage with tuneable conductance states. Ideal for in-situ training.

Profile Vendor Family Width
ibm_ecram IBM ECRAM-AnalogAI 8-bit
samsung_pcram Samsung PCRAM 8-bit
stanford_ecram Stanford ECRAM-Research 8-bit

Wafer-Scale (3 profiles)

Physics: Entire silicon wafers used as single chips. Cerebras WSE-3 has 4 trillion transistors, 900,000 cores. Tesla Dojo uses custom die-to-die interconnect for distributed neural training.

Profile Vendor Family Width Freq
cerebras_wse3_ws Cerebras WSE-3-WS 16-bit 1 GHz
tesla_dojo3 Tesla Dojo-3 16-bit 2 GHz
tachyum_prodigy Tachyum Prodigy-2nm 16-bit 5.5 GHz

Analog Mixed-Signal (2 profiles)

Physics: Compute-at-sensor architectures. Aspinity AML100 performs analog feature extraction before ADC, eliminating >90% of data movement. Renesas AnalogAI integrates analog MAC arrays.

Profile Vendor Family Width
aspinity_aml100 Aspinity AML100 8-bit
renesas_analog_ai Renesas AnalogAI 8-bit

Memory-Centric and Quantum-Inspired Platforms

RRAM / Memristive Crossbar (3 profiles)

Physics: Resistive RAM crossbar arrays perform in-situ matrix-vector multiplication using Ohm's law (V=IR) and Kirchhoff's current law. Weebit Nano demonstrates 200 TOPS/W, licensed to TI and Onsemi.

Profile Vendor Family Width
weebit_reram Weebit Nano ReRAM-ACiM 8-bit
crossbar_rram Crossbar ReRAM-1T1R 8-bit
adesto_cbram Adesto CBRAM 8-bit

Compilation notes: - Weights stored as conductance states (analog) - Use drift compensation (§32) for long-term reliability - 8-bit is typical due to conductance resolution limits

SRAM Compute-in-Memory (2 profiles)

Physics: Digital CIM using standard 6T SRAM cells modified for in-memory Boolean/MAC operations. TSMC and Samsung offer foundry- qualified CIM macro IP at N7 and SF3 nodes.

Profile Vendor Family Width Freq
tsmc_cim_n7 TSMC CIM-N7 8-bit 1 GHz
samsung_cim_sf3 Samsung CIM-SF3 8-bit 900 MHz

Cryogenic CMOS (2 profiles)

Physics: Standard CMOS operated at 4K for quantum chip control. Device physics change significantly at cryogenic temperatures (threshold shift, carrier freeze-out, reduced leakage).

Profile Vendor Family Width Freq
intel_horse_ridge Intel Horse-Ridge-II 16-bit 6 GHz
google_cryo_ctrl Google Cryo-Controller 16-bit 4 GHz

DNA / Molecular (2 profiles)

Physics: DNA base-pair gated logic and perovskite-DNA hybrid synaptic devices. Microsoft demonstrated enzymatic DNA synthesis for archival compute. ASU achieved CMOS-DNA integration.

Profile Vendor Family Width
microsoft_dna_store Microsoft DNA-Storage 8-bit
asu_dna_perovskite ASU DNA-Perovskite 8-bit

Quantum Neuromorphic (2 profiles)

Physics: Quantum reservoir computing and quantum SNNs using superconducting transmon qubits (IBM) or trapped-ion all-to-all connectivity (IonQ).

Profile Vendor Family Width
ibm_qnn IBM Quantum-NN 16-bit
ionq_trapped_ion IonQ Trapped-Ion-QNN 16-bit

Interconnect, Acoustic, Fluidic, and Space Platforms

Optical Interconnect / CPO (2 profiles)

Physics: Silicon photonic I/O chiplets replacing electrical interconnect with optical. Ayar Labs TeraPHY delivers 8 Tbps bidirectional, UCIe-compatible. Enables rack-scale SNN.

Profile Vendor Family Width Freq
ayar_teraphy Ayar Labs TeraPHY 16-bit 25 GHz
intel_cpo Intel CPO 16-bit 20 GHz

Acoustic / Phononic (2 profiles)

Physics: Acoustic wave reservoir computing using MEMS resonator arrays. Mechanical nonlinearity provides natural activation functions. Zero digital power consumption during inference.

Profile Vendor Family Width
mit_phononic MIT Phononic-NN 8-bit
caltech_mems_nn Caltech MEMS-NN 8-bit

Fluidic / Microfluidic (2 profiles)

Physics: Droplet-based and pressure-driven bistable logic gates for chemical/biological neural computation. Lab-on-chip applications for in-vivo diagnostic neural inference.

Profile Vendor Family Width
stanford_microfluidic Stanford µFluidic-NN 8-bit
eth_fluidic_logic ETH Zurich Fluidic-Logic 8-bit

Space-Qualified (4 profiles)

Physics: Radiation-hardened processors qualified for TID (Total Ionizing Dose) and SEE (Single Event Effects) tolerance. Deployed on ISS, Mars rovers, and deep-space missions.

Profile Vendor Family Width Freq
bae_rad750_sq BAE Systems RAD750 32-bit 200 MHz
seakr_sbc SEAKR SBC-SpaceAI 16-bit 400 MHz
vorago_va10820 Vorago VA10820 16-bit 100 MHz
frontgrade_leon5 Frontgrade LEON5-FT 32-bit 250 MHz

Compilation notes: - Use TMR (Triple Modular Redundancy) via §7 SEU hardening - Combine with §50 fault tree for DO-254 Level A certification - Use supply chain risk scorer (§36) for ITAR compliance

Python
from sc_neurocore.compiler.platforms import get_profile
from sc_neurocore.compiler.intelligence import (
    score_supply_chain_risk, generate_fault_tree,
)

p = get_profile("bae_rad750_sq")
risk = score_supply_chain_risk("bae_rad750_sq")
ft = generate_fault_tree("sc_lif", {"v": "-(v)/tau + I"})
print(f"Risk: {risk.overall_risk}")
print(f"Fault tree events: {len(ft.basic_events)}")

Sovereign, Organic, and Magnonic Platforms

Magnonic / Skyrmion (3 profiles)

Physics: Magnetic skyrmions are topologically protected spin textures that behave as quasi-particles. Their nonlinear dynamics, ultra-low switching energy (~10⁻²⁰ J), and emergent collective behaviour make them ideal reservoir computing substrates. Spin-wave (magnon) interference provides natural nonlinear activation without transistors.

Profile Vendor Family Width
tum_skyrmion TU Munich SkyANN-v1 8-bit
kaist_spinwave KAIST SpinWave-RC 8-bit
imec_mtj_reservoir imec MTJ-Reservoir 8-bit

Compilation notes: - Reservoir computing mode: only readout layer trained - Use 8-bit widths (sub-fJ switching limits precision) - Ideal for temporal signal processing (EEG, vibration)

Python
from sc_neurocore.compiler.platforms import get_profile

p = get_profile("tum_skyrmion")
print(f"Class: {p.platform_class}")  # magnonic

Organic Bioelectronic (2 profiles)

Physics: Organic Electrochemical Transistors (OECTs) use mixed ionic/electronic conduction in PEDOT:PSS to create artificial synapses that operate in aqueous environments. Enables direct neural tissue interfacing for in-vivo bioelectronic medicine.

Profile Vendor Family Width
cambridge_oect Cambridge OECT-Synapse 8-bit
linkoping_organic Linköping Organic-NN 8-bit

Compilation notes: - Output is stimulation/recording protocol, not RTL - Wet computing: ionic time constants (100 ms – 10 s) - Biocompatible — no heavy metals - Use HIL calibration (§62) for analog drift compensation

RISC-V Sovereign AI (5 profiles)

Physics: Open-ISA RISC-V processors with vector AI extensions. No ITAR/EAR restrictions. Enables data-sovereign and export-control-free deployment for government, defence, and critical infrastructure.

Profile Vendor Family Width Freq
sifive_x280_ai SiFive X280-AI 16-bit 2 GHz
esperanto_et_soc Esperanto ET-SoC-1 8-bit 1 GHz
ventana_veyron_ai Ventana Veyron-V2 16-bit 3.6 GHz
tenstorrent_ascalon Tenstorrent Ascalon 16-bit 4 GHz
andes_ax45mpv Andes AX45MPV 16-bit 1.5 GHz

Compilation notes: - Open ISA: no license fees or export restrictions - Use UCIe protocol mapper (§64) for chiplet-based designs - Combine with SBOM generator (§61) for EU CRA compliance - Supply chain risk score (§36) will be LOW for all RISC-V

Python
from sc_neurocore.compiler.platforms import get_profile
from sc_neurocore.compiler.intelligence import score_supply_chain_risk

p = get_profile("sifive_x280_ai")
risk = score_supply_chain_risk("sifive_x280_ai")
print(f"Open ISA — Risk: {risk.overall_risk}")

Custom Profile Registration

Method 1: TOML File (Zero Code Changes)

Create a .toml file with your custom profiles:

TOML
# my_lab_profiles.toml
[[profile]]
name = "my_custom_asic"
vendor = "UniversityLab"
family = "ASIC-v1"
platform_class = "asic"
data_width = 24
fraction = 12
overflow = "saturate"
rounding = "nearest"
max_freq_mhz = 500
notes = "Custom 24-bit ASIC for cortical simulation."

[[profile]]
name = "my_fpga_board"
vendor = "UniversityLab"
family = "Custom-FPGA"
platform_class = "fpga"
data_width = 16
fraction = 8
overflow = "wrap"
rounding = "truncate"
dsp_block = "DSP48E2"
dsp_mult_a = 18
dsp_mult_b = 27
max_freq_mhz = 250
notes = "Custom FPGA board with Xilinx UltraScale+."
Python
from sc_neurocore.compiler.intelligence import load_profiles_from_toml

loaded = load_profiles_from_toml("my_lab_profiles.toml")
print(f"Loaded: {loaded}")  # ['my_custom_asic', 'my_fpga_board']

Method 2: Runtime Discovery Hook (Vendor SDK)

Python
from sc_neurocore.compiler.intelligence import (
    register_platform_hook, discover_platforms,
)
from sc_neurocore.compiler.platforms import HardwareProfile

def vendor_discovery():
    """Auto-detect connected hardware and return profiles."""
    return [HardwareProfile(
        name="detected_board", vendor="AutoDetect",
        family="Board-v1", platform_class="fpga",
        data_width=16, fraction=8,
        overflow="saturate", rounding="nearest",
    )]

register_platform_hook(vendor_discovery)
discovered = discover_platforms()

Method 3: Auto-Construct from Spec Sheet

Python
from sc_neurocore.compiler.platforms import HardwareProfile

# Any future chip — one function call:
p = HardwareProfile.from_constraints(
    "my_2030_chip",
    vendor="FutureVendor",
    platform_class="custom",
    max_power_budget_mw=5,
    min_precision_bits=8,
)
# Automatically registered and ready to compile against

Frontier Paradigm Platforms

Wetware / Biological (2 profiles)

Physics: Living organoid co-processors interfaced via high-density Multi-Electrode Arrays (MEAs). FinalSpark uses living spherical brain organoids for closed-loop biocomputing. Cortical Labs (DishBrain) demonstrates embodied intelligence via active biological neural cultures.

Profile Vendor Family Width
finalspark_neuroplatform FinalSpark Neuroplatform 8-bit
cortical_labs_dishbrain Cortical Labs DishBrain 8-bit

Compilation notes: - Uses the map_wetware_mea (§79) intelligence feature to translate SNN topology into spatio-temporal stimulations.

Molecular / Chemical (2 profiles)

Physics: Computation and storage within synthetic DNA base pairs and enzymatic reactions. Biomemory maps ultra-high-density data (such as billion-parameter network weights) into physical DNA storage at near-zero static energy. Catalog utilizes parallel search within DNA liquid solutions.

Profile Vendor Family Width
biomemory_dna Biomemory DNA-Storage 8-bit
catalog_dna_compute Catalog Shannon 8-bit

Reversible / Adiabatic (2 profiles)

Physics: Operating at the Landauer limit of energy dissipation. Logic gates (like Toffoli and Fredkin) preserve information perfectly, allowing energy to be recovered rather than dissipated as heat. Requires multi-phase trapezoidal resonant clocking (§82).

Profile Vendor Family Width
superconducting_aqfp Yokohama Univ AQFP 16-bit
scrl_logic Generic SCRL 16-bit

Microfluidic / Mechanical (2 profiles)

Physics: Fluid dynamics and nonlinear mechanical oscillators acting as physical computational substrates. Nanofluidic 2D ionic channels physically emulate biological ion exchange using water/ion flow.

Profile Vendor Family Width
nanofluidic_logic EPFL Ion-Channel 8-bit
mems_neuromorphic Generic MEMS-Resonator 8-bit

Representative platform groups documented in this guide:

ID Class Name Profiles Description Target
1 fpga 27 Traditional FPGAs (Xilinx/Intel) Production
2 asic 13 Custom silicon Production
3 neuromorphic 11 SNN chips (Loihi, TrueNorth) Production
4 pim 8 Processing-in-memory Production
5 quantum 6 Superconducting/Ion/Optical Research
6 optical 6 Silicon photonics Pre-production
7 analog 6 Continuous-time analog Research
8 memristive 6 Crossbar arrays Research
9 emerging 3 Hybrid/novel Research
10 superconducting 3 SFQ/AQFP at 4K Research
11 spintronic 2 STT/SOT-MRAM Pre-production
12 ferroelectric 2 FeFET/FeRAM Pre-production
13 cgra 3 Reconfigurable array Production
27 fluidic 2 Microfluidic logic Research
28 space_qualified 4 Rad-hard processors Deployed
29 magnonic 3 Skyrmion/spin-wave Research
30 organic_bioelectronic 2 OECT wet computing Research
31 risc_v_sovereign 5 Open ISA AI cores Production
32 thermodynamic 2 EBM thermal equilibration Research
33 probabilistic 2 p-Bit sMTJ Research
34 polariton 2 Bose-Einstein condensate Research
35 metamaterial 2 Passive wave propagation Research
36 wetware 2 Living organoids (MEA) Research
37 molecular 2 DNA / Enzymatic compute Research
38 reversible 2 Adiabatic / Landauer limit Research
39 microfluidic 2 Nanofluidic / MEMS Research
Total 191

Further Reading