NEURA PARSE
EN
NODERIQ hardware cards

EV1 / Integration and test guide

Turn a hardware profile into a reviewable test.

Start with the dimensional and electrical configuration, run a named fixture and keep the measurements. The same criteria feed the product guides and the offline checklist checker.

Follow the bench procedureDownload the integration guide

EV1 is an engineering evaluation record. Bench criteria are not published test results, component peaks are not board benchmarks, and a completed checklist does not authorise physical operation.

Bench procedure

One configuration. One traceable record.

  1. 01

    Freeze the configuration

    Identify the card revision, sample, processor or module, firmware, harness and workload. Confirm the electrical release and approved pin map before energising hardware. Record the operator, instrument identifiers and calibration dates.

  2. 02

    Check the mechanical and power path

    Measure the PCB and mounting pattern against the dimensional source. Inspect fastener clearance, connector access and strain relief. Use one approved supply path, verify rail extrema and check for unintended USB or peripheral backfeed.

  3. 03

    Exercise the named fixture

    Run the card-specific interface fixture and preserve the raw capture. Keep source configuration, units and errors alongside each result. Neural-model throughput, analog accuracy and bus receipt describe different measurements.

  4. 04

    Review the evidence together

    Attach the raw-record identifiers and SHA-256 hashes, then check completeness and numeric criteria. An engineer reviews the physical configuration and signs the result separately. The checklist does not authorise machine operation or create certification.

Card-specific screening values

Keep the target and the result separate.

Use the stated bench ambient range and configured workload. Surface screening values are initial engineering criteria; they are not touch-safety limits, environmental ratings or a certified thermal envelope.

NODERIQ Core C10045 W / 30 min / 20–25 °C

Configured Orin NX mode at or below 25 W; one camera or host data source and local storage. Record the exact module, JetPack image and workload.

Power and thermal screening

Input range
9–24 V
Regulated rail
4.75–5.25 V
Maximum steady input power
45 W
Maximum measured surface
70 °C
Minimum workload duration
30 min
Bench ambient range
20–25 °C

Wired network link

Minimum negotiated link / Mbit/s
1000
Minimum transmitted packets
1000
Maximum lost packets
0

USB cold-start enumeration

Minimum cold starts
3
Maximum enumeration failures
0
Download the blank measurement record
NODERIQ Vision V8010 W / 30 min / 20–25 °C

A fixed 640 × 480 camera capture fixture; record sensor, pixel format, exposure and compiled firmware. Evaluate any neural model in a separate named workload.

Power and thermal screening

Input range
4.75–5.25 V
Regulated rail
3.135–3.465 V
Maximum steady input power
10 W
Maximum measured surface
65 °C
Minimum workload duration
30 min
Bench ambient range
20–25 °C

USB cold-start enumeration

Minimum cold starts
3
Maximum enumeration failures
0

Camera acquisition

Image width / pixels
640
Image height / pixels
480
Minimum captured frames
1000
Maximum missing frames
0
Minimum capture rate / frames/s
10
Download the blank measurement record
NODERIQ Link L704 W / 30 min / 20–25 °C

Two independent CAN fixtures at 500 kbit/s arbitration and a 2 Mbit/s data-phase target; a separate RS-485 loopback fixture. Record transceiver, termination and firmware.

Power and thermal screening

Input range
4.75–5.25 V
Regulated rail
3.135–3.465 V
Maximum steady input power
4 W
Maximum measured surface
60 °C
Minimum workload duration
30 min
Bench ambient range
20–25 °C

Independent CAN buses

Independent physical buses
2
Minimum frames per bus
10000
Arbitration rate / kbit/s
500
Data-phase rate / kbit/s
2000
Maximum error frames
0

Serial loopback

Serial baud rate
115200
Minimum transmitted messages
10000
Payload length / bytes
64
Maximum payload mismatches
0
Download the blank measurement record
NODERIQ Sense S501 W / 30 min / 20–25 °C

Four-channel acquisition at a 4 kSPS evaluation setting, gain 1, a traceable 0.5 V differential stimulus within the approved common-mode range, and a stationary IMU fixture.

Power and thermal screening

Input range
4.75–5.25 V
Regulated rail
3.135–3.465 V
Maximum steady input power
1 W
Maximum measured surface
55 °C
Minimum workload duration
30 min
Bench ambient range
20–25 °C

Analog reference acquisition

Simultaneous analog channels
4
Sample rate / samples/s
4000
Analog gain
1
Differential reference / V
0.5
Reference tolerance / V
0.001
Maximum absolute error / mV
2
Common-mode reference / V
1.65
Common-mode tolerance / V
0.1

Stationary inertial fixture

Minimum stationary duration / s
60
Minimum acceleration magnitude / g
0.9
Maximum acceleration magnitude / g
1.1
Maximum mean gyro magnitude / degrees/s
2
Download the blank measurement record

Fixture methods

Record what actually happened.

Mechanical fit

Use callipers and the 1:1 outline or millimetre DXF. Record width, height, thickness, maximum hole-centre error and total assembled height. Inspect every fastener and mating connector with the intended enclosure and harness.

Record to retain

Measured dimensions, mounting photograph and enclosure/harness revision.

Power rails and backfeed

With the approved assembly and supply, measure input and regulated-rail minimum/maximum values during start-up and the named workload. Record steady input power and test disconnected host/peripheral supply paths for unintended backfeed.

Record to retain

Power trace, instrument range, probe points, supply limit and fitted configuration.

Sustained thermal load

Run the exact workload for the specified interval in the stated bench ambient range. Record the hottest measured surface, configured compute mode, cooling arrangement, resets and thermal-throttle flags. These screening conditions do not define an outdoor operating rating.

Record to retain

Timestamped thermal trace, ambient measurement, surface location and workload version.

Wired network link

Connect to a known-good wired host or switch and record the negotiated link rate. Send the defined packet fixture, count loss and preserve the interface statistics. Link negotiation is not an application-throughput benchmark.

Record to retain

Host configuration, link negotiation and packet-count capture.

USB cold-start enumeration

Repeat cold starts with the approved cable and host. Record the enumerated device and the actual firmware-provided identifiers; do not assume that an unreleased image implements a USB interface. Count enumeration failures explicitly.

Record to retain

Host USB log, firmware image identifier, cable and cold-start count.

Camera acquisition

Use one approved sensor, ribbon and firmware pipeline. Fix image size, pixel format and exposure, then count received and missing frames over the named capture. Evaluate a neural model separately with its model version, quantisation and dataset.

Record to retain

Camera part, ribbon, exposure, pixel format, frame count and capture timing.

Independent CAN buses

Use two independent CAN fixtures with deliberate termination and reference-domain wiring. Record arbitration and data-phase rates, sent/received frames and error frames separately for each bus. A repeated bus identifier cannot stand in for a second physical interface.

Record to retain

Two bus traces, transceiver configuration, termination and firmware revision.

Serial loopback

Use the released RS-485 harness, direction control and termination. Send the fixed-length loopback messages at the specified baud rate, compare every returned payload and record missing or mismatched messages.

Record to retain

Serial capture, baud, message size, sent/received counts and mismatches.

Analog reference acquisition

Use the electrically released low-voltage analog front end and a traceable reference. Set the declared gain, common-mode voltage and sample rate. Exercise all four channels simultaneously and record the largest absolute error; ADC word length is not calibrated accuracy.

Record to retain

Reference-source and DMM records, input/common-mode conditions and per-channel capture.

Stationary inertial fixture

Fix the complete assembly in a documented stationary orientation away from vibration. Record acquisition duration, acceleration-vector magnitude and the largest absolute mean gyro reading. Keep the mounting frame and temperature with the capture.

Record to retain

Orientation photograph, temperature, duration and raw inertial samples.

Physical authority boundary

Keep actuator power disconnected during the fixture and retain the machine's independent interlock. Confirm that reconnecting a data link cannot automatically resume physical action. Identify the separate controller and review record that govern any later machine test.

Record to retain

Fixture boundary record, controller/interlock identity and supervised sign-off path.

Local evidence review

Check completeness before the engineering review.

Blank values remain missing. Synthetic fixtures remain examples. A valid record needs matching card identity, configuration, current instrument metadata, raw-record identifiers and SHA-256 hashes.

Select the JSON record you filled from the matching blank template. The check runs locally in your browser; no file is sent to a server.

No measurement record has been checked.

Use the same checker offline.

The standalone checker uses Node.js 20 or later and the same evaluation logic as this page. A ready-for-review result is not a physical acceptance certificate.

node validate-evidence.mjs core-c100-specification.json measurements.json
Download the offline checker

Primary-source research / 23 October 2026

Current components, explicit availability.

Orin NX, industrial application processors and MCU-local AI serve different integration roles. Announced Nano 2 and Thor T2000/T3000 products have future availability; they do not establish EV1 carrier compatibility.

Raspberry Pi

AI HAT+ 2

Hailo-10H and dedicated memory provide a current Pi 5 acceleration alternative; HAT and Orin connectors are different.

NXP

i.MX 95 processor family

Industrial application, real-time and safety domains are useful references; SoC facilities do not certify a custom board.

STMicroelectronics

STM32N657X0

V80 basis: 14 × 14 mm VFBGA264 package, Cortex-M55, on-chip SRAM, camera pipeline and NPU. Board performance requires workload measurements.

STMicroelectronics

STM32H753VI

L70 basis: 14 × 14 mm LQFP100 package and two CAN FD controllers; external transceivers and isolation still require design and validation.

STMicroelectronics

STM32U575CG

S50 basis: 7 × 7 mm UFQFPN48 option, USB full speed and CAN FD; actual pin multiplexing must be resolved in the electrical release.

Texas Instruments

ADS131M04 simultaneous ADC

Four simultaneous 24-bit channels. EV1 tests use 4 kSPS rather than asserting the component maximum as board throughput or accuracy.

Bosch Sensortec

BMI088 inertial sensor

3 × 4.5 × 0.95 mm inertial package; vibration and temperature behavior must be measured in the final mounting arrangement.

Download the complete family packDiscuss the evaluation scope