HLS4ML
Advisor
HARDWARE ACCELERATION EXPERT
Estimate DSPs, LUTs, latency · get an optimal config
① FPGA Platform
PYNQ-Z1
220 DSP · 53K LUT
PYNQ-Z2
220 DSP · 53K LUT
Custom
define manually
Available DSPs
Available LUTs
Clock Frequency (MHz)
200
MHz
② Network Architecture (Dense layers)
Add a layer (number of neurons)
+ Add
Reset
Click a layer node to remove it
③ ap_fixed Precision
🏋️ Weights
W (total bits)
I (integer bits)
ap_fixed<8,1>
⚡ Activations
W (total bits)
I (integer bits)
ap_fixed<8,4>
Quantization Preset
— Choose a preset —
HLS4ML Default (ap_fixed<16,6>)
QKeras aggressive 6-bit
QKeras standard 8-bit
Intermediate 12-bit
Float32 (reference)
④ ReuseFactor
1
RF=1 → Max parallel (min latency)
RF=N → Min DSPs (max latency)
⑤ Pruning & Sparsity
Sparsity (fraction of weights = 0)
0%
0% → no pruning
75% → typical QKeras
90%+ → very aggressive
⑥ Latency Target
Max desired latency (µs)
Strategy mode
Latency
Resource
⚡
Configure the parameters
Results will appear here in real time