A dexterous hand has a problem that most robotic systems do not: it needs absolute 3D position feedback from a dozen joints sitting inside a few cubic centimetres, running on battery or bus power, with an encoder the size of a grain of rice. The MLX90393 from Melexis has been a common answer. The CONNTEK KTH5701AQ2DNE is a drop-in-class alternative that changes the trade-off in three places worth knowing about.
Short version: if your joint design needs large magnetic range (a strong magnet or a wide air gap), a hardware angle output without MCU computation, or a 105°C industrial temperature grade, the KTH5701AQ2DNE is the better fit. If you need the highest angular resolution in a very weak field, the MLX90393 still leads on raw sensitivity.
Where the two parts actually differ
Both are 3D Hall sensors in a QFN 3×3 mm package with I²C and SPI. That is where the similarity ends.
| Parameter | KTH5701AQ2DNE | MLX90393 |
|---|---|---|
| Manufacturer | CONNTEK (China) | Melexis (Belgium) |
| Package | QFN 3×3-16L | QFN 3×3-16L |
| XY linear range | ±130 mT | ±50 mT |
| Z linear range | ±80 mT | ±50 mT |
| Sensitivity XY | 65.5 LSB/mT | up to ~6211 LSB/mT |
| Sensitivity Z | 102 LSB/mT | up to ~3406 LSB/mT |
| ADC | 16-bit fixed | 17-bit core, 16-bit output |
| Angle output | Hardware CORDIC, XY/XZ/YZ planes | None, computed in MCU |
| Analog VDD | 2.8 – 5.5 V | 2.2 – 3.6 V |
| IO supply | down to 1.8 V | down to 1.65 V |
| Temperature grade | −40 to +105 °C (industrial) | −40 to +85 °C (E grade) |
| Idle current | 1.4 µA | 2.4 µA |
| Wake-standby current | 2.4 µA | 43 µA |
| ESD (HBM) | ±5 kV | 2.5 kV |
| SPI clock | up to 5 MHz | up to 10 MHz |
| Key detection pin | Dedicated BUTT_OUT / TRIG | INT/TRIG shared, no dedicated pin |
Sources: CONNTEK KTH5701 series datasheet; CONNTEK original comparison table for KTH5701 vs MLX90393.
The four differences that matter in a dexterous hand
1. Magnetic range decides your mechanical design. A dexterous hand joint is small, so the magnet is small, so the field at the sensor is weak — which sounds like it favours the higher-sensitivity MLX90393. In practice the opposite is often true: assembly tolerances in a finger joint are loose, the air gap varies, and a ±50 mT ceiling clips on the strong side of the travel. The KTH5701AQ2DNE's ±130 mT XY range means the sensor keeps reading linearly even when the magnet is closer than the drawing specifies.
2. Hardware angle output removes MCU load. The KTH5701AQ2DNE has an on-chip CORDIC engine that outputs angle directly in XY, XZ and YZ planes. The MLX90393 returns raw magnetic field values only, so the host MCU must run the arctangent math for every joint, every cycle. In a hand with 12 to 20 joints sharing one MCU, that is real CPU time and real firmware.
3. Standby current sets battery life. Both parts idle low. The gap is in wake-standby: 2.4 µA versus 43 µA. On a device that wakes, measures and sleeps hundreds of times per second, that 18× difference dominates the average current budget.
4. Temperature grade and ESD headroom. The KTH5701AQ2DNE is rated to +105 °C, and its AQ2 industrial grade covers that range. The MLX90393's E grade tops out at +85 °C. For a hand mounted near a motor or in an enclosed actuator, the extra 20 °C is margin you will want. The ±5 kV HBM rating is also double, which matters when the sensor sits at the end of a flex cable on a moving assembly.
What you give up
Be clear about the other direction too. The MLX90393 wins on sensitivity — its programmable gain reaches roughly 6211 LSB/mT on XY, which means finer resolution in very weak fields. It also wins on SPI speed (10 MHz versus 5 MHz) and on having four selectable I²C addresses on the address pins, where the KTH5701AQ2DNE requires A0/A1 grounded in SPI mode.
If your joint uses a strong magnet at a fixed, tight air gap and you want the maximum number of counts per degree, the Melexis part is still the sharpest tool.
Selection checklist
Ask these before choosing:
- Magnet strength and air gap tolerance — loose tolerance or a strong magnet points to KTH5701AQ2DNE (±130 mT); a small weak magnet points to MLX90393.
- Who computes the angle — if you want the sensor to hand you degrees, KTH5701AQ2DNE.
- Ambient temperature at the joint — above 85 °C, KTH5701AQ2DNE.
- Average current budget — high wake/sleep frequency, KTH5701AQ2DNE.
- Maximum angular resolution in a weak field — MLX90393.
Frequently asked questions
Is the KTH5701AQ2DNE pin-compatible with the MLX90393?
Both are QFN 3×3-16L, so the footprint matches, but the pin functions are not identical. The KTH5701AQ2DNE uses dedicated A0/A1 address pins that must be grounded in SPI mode and provides a separate BUTT_OUT/TRIG pin for key detection, where the MLX90393 shares INT and TRIG. Treat it as a same-footprint redesign, not a solder-and-go swap.
Can I get angle data without writing arctangent code?
Yes, on the KTH5701AQ2DNE. Its CORDIC engine outputs angle and magnitude in the XY, XZ or YZ plane directly, in a TABZ readback format. The MLX90393 returns raw TXYZ values and expects the MCU to compute the angle.
Which part is better for a battery-powered prosthetic hand?
The KTH5701AQ2DNE, on current budget. Its wake-standby current is 2.4 µA against 43 µA for the MLX90393, and it also idles lower (1.4 µA vs 2.4 µA). At high duty cycles that difference exceeds the modest per-conversion advantage the MLX90393 has.
Do both parts work at 1.8 V logic?
Yes. The KTH5701AQ2DNE's IO supply goes down to 1.8 V while its analog supply runs 2.8 to 5.5 V. The MLX90393's IO goes slightly lower at 1.65 V, but its analog supply only reaches 3.6 V — so for a 5 V rail design, the KTH5701AQ2DNE is the one that fits.
Sourcing note
Both parts are available through Mandu with full original documentation. The KTH5701AQ2DNE ships from CONNTEK's standard product line; send your target quantity and annual volume and we will return pricing, availability and the datasheet package together.