GaN　HEMT　switches　has　become　more　and　more　important　in　RF　front-end.　This　article　proposes　a　circuit-based　neuro-SM　technique　for　the　small-signal　modeling　of　multi-gate　GaN　HEMT　switches　for　the　first　time.　A　general　equivalent　circuit　model　for　multi-gate　switch　HEMT　is　proposed　using　the　cascaded　single-gate　switch　HEMT　model　to　represent　the　tendency　rather　than　the　exact　behaviors　of　the　small-signal　responses.　We　utilize　this　proposed　equivalent　circuit　model　as　the　coarse　model　and　propose　a　novel　circuit-based　neuro-SM　modeling　technique.　Neural　networks　are　incorporated　to　learn　the　difference　between　the　coarse　model　and　the　fine　device　data,　improving　the　neuro-SM　model　accuracy.　After　being　trained,　the　obtained　neuro-SM　model　can　be　used　for　high-level　circuit　design　to　increase　the　speed　and　accuracy　of　circuit　design.　Compared　to　the　existing　modeling　techniques　for　multi-gate　switches,　the　proposed　neuro-SM　achieves　better　model　accuracy.　Examples　of　a　dual-gate　GaN　HEMT　switch　and　a　triple-gate　GaN　HEMT　switch　have　been　examined.