
Automotive long rangeRadar in Arm based SoC
The customer goal was to have a long
range Radar supporting multiple channels in
a virtual array, giving it a high resolution
imaging capability. This needed to be
developed and qualified in less than 2 years
to meet the start of the production date for
a new vehicle model where this Radar was a
key differentiating feature.
Customer Challenge
Complex Radar processing SoC
An automotive Tier 1 approached EnSilica
with a requirement of a complex Radar
processing SoC developed using an Arm
based FPGA. The Tier 1 had intended to use an
ASSP in their OEMs new model, but its
availability was delayed and would not be
ready in time for the OEM’s start of production
date.
The challenge was to develop, verify, and for
the Tier 1 to road test and qualify this state-ofthe-
art Radar SoC using one of the smaller
AMD UltraScale+ FPGA. To complicate matters
this design had to meet a stringent power
budget.
Solution
New Radar system that brings customer advantage
EnSilica and the Tier 1 collaborated on system level
design and specification developing MATLAB and C
system models.
First prototype was required for
collecting data for field trial and off-line processing,
a data-logging system interfacing to four 77GHz RF
ICs each supporting 4 Rx channels and 3 Tx
channels via MIPI. The FPGA performed a 2D FFT and
logged the data. The system was delivered to the
customer in less than 8 weeks, taking advantage of
our prior art and IP.
EnSilica was responsible for all the FPGA
programmable logic (PL) design in RTL and lowlevel
software drivers. EnSilica's IP was customised
to fit the needs of most of the signal processing
path and tracking functions.
EnSilica was responsible for the verification and functional
safety aspects of the PL, whilst also supporting the
Tier 1 to develop the system so it met the
requirements of ISO26262 ASIL-B.
As well as the FPGA implementation, EnSilica
developed a bit-exact version to run on a NVIDIA
GPU farm accelerated using CUDA®. This ran several
times faster than real-time allowing off-line data
captured during field trials to be quickly processed
Using the ARM compiler, simulator and other tools
we were able to co-simulate the hardware/
software split for object identification and tracking,
thereby verifying this capability against
specification. Advanced debug features in
Coresight allowed for breakpoints to be set in a
combination of hardware and software domains,
enabling complex scenarios to be debugged with
ease.
Using EnSilica's hardware accelerator IP cores the
design and verification was complete within 12
months allowing plenty of time for field trials and
automotive qualification of the model before the
release of the new vehicle model by the OEM. The
new Radar system had enhanced performance
and was highly differentiated from their
competitors. The customer would not have been
able to deliver this differentiation if they had used
the Radar ASSP as originally planned.
Application Area
Automotive
Functional Safety
Artificial Intelligence
EnSilica IP
Multiple FFT processors cores
Multiple Kalman filter trackers
Floating point matrix accelerator
Maximum Likelihood estimator core
Power spectral density accelerator
eSi-Crypto AES and SHA
3rd Party IP
Arm Cortex-R5F CPU
Arm Cortex-A53 CPU
Arm Mali-400 CPU
4 channels of MIPI Rx controller for interfacing to the 77GHz RF ICs
EnSilica owns a library of Radar processing hardware accelerators, that has been licensed to chip makers for many years.
Fast Fourier Transforms (FFT)
Constant false alarm rate (CFAR) processor
Kalman Tracking Filter tracker
Real valued matrix IP – add/sub/div determinant, RQ decomposition, Back substitution, Cholesky decomposition
Maximum Likelihood and Super resolution IP -SVD of non-square matrix, MUSIC eigenvector extraction, MinNorm vector extraction,Diagonal loading and Capon matrix extraction

