News & Analysis
Megachips to Launch DSP-Based Sensor Fusion IC
The battle over sensor fusion chips heats up
OSAKA, Japan — The emerging sensor-fusion controller market for smartphones and wearable devices is about to meet a new competitor -- this time out of left field. Megachips, Japan’s fabless, is rolling out a motion engine and sensor hub chip, dubbed Frizz.
Unlike existing sensor-fusion controller chips, Frizz isn’t MCU-based. (Sensor hubs usually use a 32-bit microcontroller core or similarly powerful processing device.) Instead, Megachips’ designers customized Tensilica (now Cadence)’s 32-bit DSP -- Xtensa Lx4 -- and expanded the hardware with 3-way VLIW and floating point 4-way SIMD.
Megachips’ deputy general manager Kenji Nakamura, tells EE Times, “Frizz is optimized for wearable devices. It can do high-level computations, such as matrix arithmetic processing, that had been only possible with a high-speed clock CPU. And Frizz does it at ultra-low power consumption.” Matrix arithmetic processing is ideal for running pedestrian dead reckoning (PDR) algorithms that require Kalman filters, for example.
Frizz calculates human movement via accelerometer, gyro, and magnetometer and determines the pedestrian's position by estimating speed and direction. By using Frizz in a wearable device with no GPS, for example, one can develop “the next-generation activity monitor.” Such a device can see differences in activity -- discerning running from walking -- and estimate walking speed and distance.
A further advantage with Frizz is its potential to enable “indoor navigation,” says Nakamura, when PDR is used in combination with map-matching and beacons. The power consumption of Megachips’ Frizz (running at 6.7MHz) when used for PDR, for example, is 2.4mW. In contrast, an MCU based on ARM Cortex M4F (running at 80MHz) consumes 38mW, he says. Frizz is specifically designed for smartphones and wearable devices in fitness/health care applications.
Motion co-processors
Today, Apple is already using such motion co-processors as Apple M7 and M8 in mobile devices. Google is also recommending the use of a sensor hub in smartphones running Android 4.4 Kitkat or above. The idea of the sensor hub is to collect sensor data from accelerometers, gyroscopes, and compasses and offload collection and processing of sensor data from the main CPU. While the market today remains split between those inclined to use an apps processor in a smartphone to do sensor data processing and others opting for a separate sensor fusion chip, Megachips believes the market is leaning toward sensor hubs.
In sum, “No chip company today would dream of getting into the application processor market for smartphones -- an empire already conquered by Qualcomm, Samsung, and Apple.” But some companies including Megachips are looking for opportunities for sensor fusion chips, he acknowledges, because that design socket could be as big as that of apps processors in smartphones.
Unlike existing sensor-fusion controller chips, Frizz isn’t MCU-based. (Sensor hubs usually use a 32-bit microcontroller core or similarly powerful processing device.) Instead, Megachips’ designers customized Tensilica (now Cadence)’s 32-bit DSP -- Xtensa Lx4 -- and expanded the hardware with 3-way VLIW and floating point 4-way SIMD.
(Source: Megachips)
Megachips’ deputy general manager Kenji Nakamura, tells EE Times, “Frizz is optimized for wearable devices. It can do high-level computations, such as matrix arithmetic processing, that had been only possible with a high-speed clock CPU. And Frizz does it at ultra-low power consumption.” Matrix arithmetic processing is ideal for running pedestrian dead reckoning (PDR) algorithms that require Kalman filters, for example.
Frizz calculates human movement via accelerometer, gyro, and magnetometer and determines the pedestrian's position by estimating speed and direction. By using Frizz in a wearable device with no GPS, for example, one can develop “the next-generation activity monitor.” Such a device can see differences in activity -- discerning running from walking -- and estimate walking speed and distance.
A further advantage with Frizz is its potential to enable “indoor navigation,” says Nakamura, when PDR is used in combination with map-matching and beacons. The power consumption of Megachips’ Frizz (running at 6.7MHz) when used for PDR, for example, is 2.4mW. In contrast, an MCU based on ARM Cortex M4F (running at 80MHz) consumes 38mW, he says. Frizz is specifically designed for smartphones and wearable devices in fitness/health care applications.
Power consumption comparison.
(Source: Megachips)
(Source: Megachips)
Motion co-processors
Today, Apple is already using such motion co-processors as Apple M7 and M8 in mobile devices. Google is also recommending the use of a sensor hub in smartphones running Android 4.4 Kitkat or above. The idea of the sensor hub is to collect sensor data from accelerometers, gyroscopes, and compasses and offload collection and processing of sensor data from the main CPU. While the market today remains split between those inclined to use an apps processor in a smartphone to do sensor data processing and others opting for a separate sensor fusion chip, Megachips believes the market is leaning toward sensor hubs.
In sum, “No chip company today would dream of getting into the application processor market for smartphones -- an empire already conquered by Qualcomm, Samsung, and Apple.” But some companies including Megachips are looking for opportunities for sensor fusion chips, he acknowledges, because that design socket could be as big as that of apps processors in smartphones.
 |
| | | ||||||||||||||||||||||||||||||||||||
 | |          | | | |||||||||||||||||||||||||||||||||||
|
| ||||||||||||||||||||||||||||||||||||||
No comments:
Post a Comment