cmod a7 development board  (Digilent Inc)

Journal: Philosophical transactions. Series A, Mathematical, physical, and engineering sciences

Article Title: Learning automata based energy-efficient AI hardware design for IoT applications

doi: 10.1098/rsta.2019.0593

Figure Lengend Snippet: Design exploration and automation flow using an FPGA.

Article Snippet: We run the same Iris dataset benchmarks across the three platforms: software Tsetlin machine running on a Raspberry Pi 3 (featuring ARM Cortex-A53 cores with 1GB LPDDR2 memory), hardware implemented on a low-power FPGA development board (Digilent Cmod A7-35T) and finally our custom ASIC hardware.

Techniques:

Journal: Philosophical transactions. Series A, Mathematical, physical, and engineering sciences

Article Title: Learning automata based energy-efficient AI hardware design for IoT applications

doi: 10.1098/rsta.2019.0593

Figure Lengend Snippet: Results of training for ASIC synthesis in 65 nm technology and runtime-reconfigurable FPGA hardware for two different datasets: a single-class noisy xor and a multi-class binary Iris. As expected, the FPGA prototype implementations return significantly higher energy than those of ASIC.

Article Snippet: We run the same Iris dataset benchmarks across the three platforms: software Tsetlin machine running on a Raspberry Pi 3 (featuring ARM Cortex-A53 cores with 1GB LPDDR2 memory), hardware implemented on a low-power FPGA development board (Digilent Cmod A7-35T) and finally our custom ASIC hardware.

Techniques:

Journal: Philosophical transactions. Series A, Mathematical, physical, and engineering sciences

Article Title: Learning automata based energy-efficient AI hardware design for IoT applications

doi: 10.1098/rsta.2019.0593

Figure Lengend Snippet: Comparison of software (Raspberry Pi) and hardware (FPGA, ASIC) platforms for ( a ) test accuracy, ( b ) training power and ( c ) training time.

Article Snippet: We run the same Iris dataset benchmarks across the three platforms: software Tsetlin machine running on a Raspberry Pi 3 (featuring ARM Cortex-A53 cores with 1GB LPDDR2 memory), hardware implemented on a low-power FPGA development board (Digilent Cmod A7-35T) and finally our custom ASIC hardware.

Techniques: Comparison, Software