diff --git a/techlibs/ice40/synth_ice40.cc b/techlibs/ice40/synth_ice40.cc
index b66c6bf..c942126 100644
--- a/techlibs/ice40/synth_ice40.cc
+++ b/techlibs/ice40/synth_ice40.cc
@@ -273,6 +273,7 @@
 			run("opt_expr");
 			run("opt_clean");
 			if (help_mode || dsp) {
+				run("memory_dff");
 				run("techmap -map +/mul2dsp.v -map +/ice40/dsp_map.v -D DSP_A_MAXWIDTH=16 -D DSP_B_MAXWIDTH=16 "
 						"-D DSP_A_MINWIDTH=2 -D DSP_B_MINWIDTH=2 -D DSP_Y_MINWIDTH=11 "
 						"-D DSP_NAME=$__MUL16X16", "(if -dsp)");
diff --git a/techlibs/xilinx/synth_xilinx.cc b/techlibs/xilinx/synth_xilinx.cc
index f137408..6f8254b 100644
--- a/techlibs/xilinx/synth_xilinx.cc
+++ b/techlibs/xilinx/synth_xilinx.cc
@@ -341,6 +341,7 @@
 
 		if (check_label("map_dsp", "(skip if '-nodsp')")) {
 			if (!nodsp || help_mode) {
+				run("memory_dff"); // xilinx_dsp will merge registers, reserve memory port registers first
 				// NB: Xilinx multipliers are signed only
 				run("techmap -map +/mul2dsp.v -map +/xilinx/dsp_map.v -D DSP_A_MAXWIDTH=25 "
 					"-D DSP_A_MAXWIDTH_PARTIAL=18 -D DSP_B_MAXWIDTH=18 "    // Partial multipliers are intentionally