inone.Thetwoprimaryprocessorsare64-bit
partsrunningatbetween 400 and600MHz
andarebasedontheopenRISC-Vinstruction
setarchitecture(ISA).Thesearejoinedbya
co-processordubbedtheKendryte
Processing Unit, with Kendryte also being the
creator of the SoC that powers the Sipeed
module, which in turn powers the Seeed HAT.
Designed for accelerating artificial
intelligence workloads – hence the name of
theHAT – this co-processor offers a claimed
230 giga-matrix-multiplications (GMULps).
That’s not all you’ll find hiding under the
metal shield either, The module includes an
audio processing system supporting up to
eight microphones, additional accelerators forCUSTOMISATION / HOBBYTECH
A
firstglanceattheSeeedStudio
GroveAIHATrevealsa surprising
amountofhardwarefora Raspberry
Pi add-onboard.There’sa reasonforthis:it’s
not reallya RaspberryPiadd-onboard.
To clarify:theSeeedStudioGroveAIHAT
can, indeed,beconnectedtoa RaspberryPi’s
general-purposeinput/output(GPIO)justlike
any otherdevicethatadheresevenvaguelyto
the RaspberryPi’sHardwareAttachedonTop
(HAT) standard.It doesn’thavetobethough:
it’s capableofoperatingasanentirely
independentdevelopmentboard,thanksto
the SipeedMaixM1AIModuleatitsheart.
This bulkypart,coveredbyanetched
metallic shield,is effectivelythreeprocessorsREVIEW
Seeed Studio
Grove AIHAT
fastFouriertransforms(FFTs),AES encryption
andSHA256hashing.There’salso6MBof
static RAM, and an additional 2MB of SRAM
for the AI accelerator.
Seeed’s contribution to the design is the
HAT-form-factor carrier board. This includes
an impressive count of six of the company’s
Grove connectors for external hardware,
covering one digital IO, one pulse-width
modulated (PWM) IO, one I²C bus, one
UART and two analogue-to-digital (ADC)
converters, along with an on-board digital
microphone and accelerometer.Right
The Grove AI HAT
has a good layout,
with plenty of
connections for
external hardwareThe hardware side of installation
is solderless and easy, but the
software is another matterLeft
It’s a HAT, or so
says the box, but
don’t expect it all
to work on a
Raspberry Pi