diff --git a/artiq/coredevice/ttl.py b/artiq/coredevice/ttl.py
index 644e55ff6..866e884e5 100644
--- a/artiq/coredevice/ttl.py
+++ b/artiq/coredevice/ttl.py
@@ -177,7 +177,7 @@ class TTLInOut:
         self._set_sensitivity(0)
 
     @kernel
-    def gate_both_mu(self, duration):
+    def gate_both(self, duration):
         """Register both rising and falling edge events for the specified
         duration (in seconds)."""
         self._set_sensitivity(3)
@@ -196,8 +196,8 @@ class TTLInOut:
 
     @kernel
     def timestamp_mu(self):
-        """Poll the RTIO input and returns an event timestamp, according to
-        the gating.
+        """Poll the RTIO input and returns an event timestamp (in machine
+        units), according to the gating.
 
         If the gate is permanently closed, returns a negative value.
         """
diff --git a/artiq/gateware/rtio/core.py b/artiq/gateware/rtio/core.py
index b4fedcf88..08c46188f 100644
--- a/artiq/gateware/rtio/core.py
+++ b/artiq/gateware/rtio/core.py
@@ -118,8 +118,10 @@ class _OutputManager(Module):
         sequence_error = Signal()
         nop = Signal()
         self.sync.rsys += [
-            replace.eq(self.ev.timestamp == buf.timestamp[fine_ts_width:]),
-            sequence_error.eq(self.ev.timestamp < buf.timestamp[fine_ts_width:])
+            replace.eq(self.ev.timestamp[fine_ts_width:] \
+                       == buf.timestamp[fine_ts_width:]),
+            sequence_error.eq(self.ev.timestamp[fine_ts_width:] \
+                              < buf.timestamp[fine_ts_width:])
         ]
         if interface.suppress_nop:
             # disable NOP at reset: do not suppress a first write with all 0s
@@ -300,8 +302,7 @@ class _KernelCSRs(AutoCSR):
 
 
 class RTIO(Module):
-    def __init__(self, channels, clk_freq, full_ts_width=63,
-                 guard_io_cycles=20):
+    def __init__(self, channels, full_ts_width=63, guard_io_cycles=20):
         data_width = max(rtlink.get_data_width(c.interface)
                          for c in channels)
         address_width = max(rtlink.get_address_width(c.interface)
@@ -329,11 +330,15 @@ class RTIO(Module):
             self.cd_rsys.rst.eq(self.kcsrs.reset.storage)
         ]
         self.comb += self.cd_rio.clk.eq(ClockSignal("rtio"))
-        self.specials += AsyncResetSynchronizer(self.cd_rio,
-                                                self.kcsrs.reset.storage)
+        self.specials += AsyncResetSynchronizer(
+            self.cd_rio,
+            self.kcsrs.reset.storage | ResetSignal("rtio",
+                                                   allow_reset_less=True))
         self.comb += self.cd_rio_phy.clk.eq(ClockSignal("rtio"))
-        self.specials += AsyncResetSynchronizer(self.cd_rio_phy,
-                                                self.kcsrs.reset_phy.storage)
+        self.specials += AsyncResetSynchronizer(
+            self.cd_rio_phy,
+            self.kcsrs.reset_phy.storage | ResetSignal("rtio",
+                                                       allow_reset_less=True))
 
         # Managers
         self.submodules.counter = _RTIOCounter(full_ts_width - fine_ts_width)
diff --git a/artiq/gateware/rtio/phy/ttl_serdes_7series.py b/artiq/gateware/rtio/phy/ttl_serdes_7series.py
new file mode 100644
index 000000000..b2322b4f5
--- /dev/null
+++ b/artiq/gateware/rtio/phy/ttl_serdes_7series.py
@@ -0,0 +1,70 @@
+from migen.fhdl.std import *
+
+from artiq.gateware.rtio.phy import ttl_serdes_generic
+
+
+class _OSERDESE2_8X(Module):
+    def __init__(self, pad):
+        self.o = Signal(8)
+        self.t_in = Signal()
+        self.t_out = Signal()
+
+        # # #
+
+        o = self.o
+        self.specials += Instance("OSERDESE2",
+                                  p_DATA_RATE_OQ="DDR", p_DATA_RATE_TQ="BUF",
+                                  p_DATA_WIDTH=8, p_TRISTATE_WIDTH=1,
+                                  o_OQ=pad, o_TQ=self.t_out,
+                                  i_CLK=ClockSignal("rtiox4"),
+                                  i_CLKDIV=ClockSignal("rio_phy"),
+                                  i_D1=o[0], i_D2=o[1], i_D3=o[2], i_D4=o[3],
+                                  i_D5=o[4], i_D6=o[5], i_D7=o[6], i_D8=o[7],
+                                  i_TCE=1, i_OCE=1, i_RST=0,
+                                  i_T1=self.t_in)
+
+
+class _IOSERDESE2_8X(Module):
+    def __init__(self, pad):
+        self.o = Signal(8)
+        self.i = Signal(8)
+        self.oe = Signal()
+
+        # # #
+
+        pad_i = Signal()
+        pad_o = Signal()
+        i = self.i
+        self.specials += Instance("ISERDESE2", p_DATA_RATE="DDR",
+                                  p_DATA_WIDTH=8,
+                                  p_INTERFACE_TYPE="NETWORKING", p_NUM_CE=1,
+                                  o_Q1=i[7], o_Q2=i[6], o_Q3=i[5], o_Q4=i[4],
+                                  o_Q5=i[3], o_Q6=i[2], o_Q7=i[1], o_Q8=i[0],
+                                  i_D=pad_i,
+                                  i_CLK=ClockSignal("rtiox4"),
+                                  i_CLKB=~ClockSignal("rtiox4"),
+                                  i_CE1=1, i_RST=0,
+                                  i_CLKDIV=ClockSignal("rio_phy"))
+        oserdes = _OSERDESE2_8X(pad_o)
+        self.submodules += oserdes
+        self.specials += Instance("IOBUF",
+                                  i_I=pad_o, o_O=pad_i, i_T=oserdes.t_out,
+                                  io_IO=pad)
+        self.comb += [
+            oserdes.t_in.eq(~self.oe),
+            oserdes.o.eq(self.o)
+        ]
+
+
+class Output_8X(ttl_serdes_generic.Output):
+    def __init__(self, pad):
+        serdes = _OSERDESE2_8X(pad)
+        self.submodules += serdes
+        ttl_serdes_generic.Output.__init__(self, serdes)
+
+
+class Inout_8X(ttl_serdes_generic.Inout):
+    def __init__(self, pad):
+        serdes = _IOSERDESE2_8X(pad)
+        self.submodules += serdes
+        ttl_serdes_generic.Inout.__init__(self, serdes)
diff --git a/artiq/gateware/rtio/phy/ttl_serdes_generic.py b/artiq/gateware/rtio/phy/ttl_serdes_generic.py
new file mode 100644
index 000000000..2ba010226
--- /dev/null
+++ b/artiq/gateware/rtio/phy/ttl_serdes_generic.py
@@ -0,0 +1,273 @@
+from migen.fhdl.std import *
+from migen.genlib.coding import PriorityEncoder
+
+from artiq.gateware.rtio import rtlink
+
+
+def _mk_edges(w, direction):
+    l = [(1 << i) - 1 for i in range(w)]
+    if direction == "rising":
+        l = [2**w - 1 ^ x for x in l]
+    elif direction == "falling":
+        pass
+    else:
+        raise ValueError
+    return l
+
+
+class _SerdesDriver(Module):
+    def __init__(self, serdes_o, stb, data, fine_ts, override_en, override_o):
+        previous_data = Signal()
+        serdes_width = flen(serdes_o)
+        edges = Array(_mk_edges(serdes_width, "rising"))
+        edges_n = Array(_mk_edges(serdes_width, "falling"))
+        self.sync.rio_phy += [
+            If(stb, previous_data.eq(data)),
+            If(override_en,
+                serdes_o.eq(Replicate(override_o, serdes_width))
+            ).Else(
+                If(stb & ~previous_data & data,
+                    serdes_o.eq(edges[fine_ts]),
+                ).Elif(stb & previous_data & ~data,
+                    serdes_o.eq(edges_n[fine_ts]),
+                ).Else(
+                    serdes_o.eq(Replicate(previous_data, serdes_width)),
+                )
+            )
+        ]
+
+
+class Output(Module):
+    def __init__(self, serdes):
+        self.rtlink = rtlink.Interface(
+            rtlink.OInterface(1, fine_ts_width=log2_int(flen(serdes.o))))
+        self.probes = [serdes.o[-1]]
+        override_en = Signal()
+        override_o = Signal()
+        self.overrides = [override_en, override_o]
+
+        # # #
+
+        self.submodules += _SerdesDriver(
+            serdes.o,
+            self.rtlink.o.stb, self.rtlink.o.data, self.rtlink.o.fine_ts,
+            override_en, override_o)
+
+
+class Inout(Module):
+    def __init__(self, serdes):
+        serdes_width = flen(serdes.o)
+        assert flen(serdes.i) == serdes_width
+        self.rtlink = rtlink.Interface(
+            rtlink.OInterface(2, 2, fine_ts_width=log2_int(serdes_width)),
+            rtlink.IInterface(1, fine_ts_width=log2_int(serdes_width)))
+        self.probes = [serdes.i[-1], serdes.oe]
+        override_en = Signal()
+        override_o = Signal()
+        override_oe = Signal()
+        self.overrides = [override_en, override_o, override_oe]
+
+        # # #
+
+        # Output
+        self.submodules += _SerdesDriver(
+            serdes_o=serdes.o,
+            stb=self.rtlink.o.stb & (self.rtlink.o.address == 0),
+            data=self.rtlink.o.data[0],
+            fine_ts=self.rtlink.o.fine_ts,
+            override_en=override_en, override_o=override_o)
+
+        oe_k = Signal()
+        self.sync.rio_phy += [
+            If(self.rtlink.o.stb & (self.rtlink.o.address == 1),
+                oe_k.eq(self.rtlink.o.data[0])),
+            If(override_en,
+                serdes.oe.eq(override_oe)
+            ).Else(
+                serdes.oe.eq(oe_k)
+            )
+        ]
+
+        # Input
+        sensitivity = Signal(2)
+        self.sync.rio += If(self.rtlink.o.stb & (self.rtlink.o.address == 2),
+            sensitivity.eq(self.rtlink.o.data))
+
+        i = serdes.i[-1]
+        i_d = Signal()
+        self.sync.rio_phy += [
+            i_d.eq(i),
+            self.rtlink.i.stb.eq(
+                (sensitivity[0] & ( i & ~i_d)) |
+                (sensitivity[1] & (~i &  i_d))
+            ),
+            self.rtlink.i.data.eq(i),
+        ]
+
+        pe = PriorityEncoder(serdes_width)
+        self.submodules += pe
+        self.comb += pe.i.eq(serdes.i ^ Replicate(i_d, serdes_width))
+        self.sync.rio_phy += self.rtlink.i.fine_ts.eq(pe.o)
+
+
+class _FakeSerdes(Module):
+    def __init__(self):
+        self.o = Signal(8)
+        self.i = Signal(8)
+        self.oe = Signal()
+
+
+class _OutputTB(Module):
+    def __init__(self):
+        serdes = _FakeSerdes()
+        self.submodules.dut = RenameClockDomains(Output(serdes),
+                                                 {"rio_phy": "sys"})
+
+    def gen_simulation(self, selfp):
+        selfp.dut.rtlink.o.data = 1
+        selfp.dut.rtlink.o.fine_ts = 1
+        selfp.dut.rtlink.o.stb = 1
+        yield
+        selfp.dut.rtlink.o.stb = 0
+        yield
+        selfp.dut.rtlink.o.data = 0
+        selfp.dut.rtlink.o.fine_ts = 2
+        selfp.dut.rtlink.o.stb = 1
+        yield
+        yield
+        selfp.dut.rtlink.o.data = 1
+        selfp.dut.rtlink.o.fine_ts = 7
+        selfp.dut.rtlink.o.stb = 1
+        for _ in range(6):
+            # note that stb stays active; output should not change
+            yield
+
+
+class _InoutTB(Module):
+    def __init__(self):
+        self.serdes = _FakeSerdes()
+        self.submodules.dut = RenameClockDomains(Inout(self.serdes),
+                                                 {"rio_phy": "sys",
+                                                  "rio": "sys"})
+
+    def check_input(self, selfp, stb, fine_ts=None):
+        if stb != selfp.dut.rtlink.i.stb:
+            print("KO rtlink.i.stb should be {} but is {}"
+                  .format(stb, selfp.dut.rtlink.i.stb))
+        elif fine_ts is not None and fine_ts != selfp.dut.rtlink.i.fine_ts:
+            print("KO rtlink.i.fine_ts should be {} but is {}"
+                  .format(fine_ts, selfp.dut.rtlink.i.fine_ts))
+        else:
+            print("OK")
+
+    def check_output(self, selfp, data):
+        if selfp.serdes.o != data:
+            print("KO io.o should be {} but is {}".format(data, selfp.serdes.o))
+        else:
+            print("OK")
+
+    def check_output_enable(self, selfp, oe):
+        if selfp.serdes.oe != oe:
+            print("KO io.oe should be {} but is {}".format(oe, selfp.serdes.oe))
+        else:
+            print("OK")
+
+    def gen_simulation(self, selfp):
+        selfp.dut.rtlink.o.address = 2
+        selfp.dut.rtlink.o.data = 0b11
+        selfp.dut.rtlink.o.stb = 1  # set sensitivity to rising + falling
+        yield
+        selfp.dut.rtlink.o.stb = 0
+
+        self.check_output_enable(selfp, 0)
+        yield
+
+        selfp.serdes.i = 0b11111110  # rising edge at fine_ts = 1
+        yield
+        selfp.serdes.i = 0b11111111
+        yield
+        self.check_input(selfp, stb=1, fine_ts=1)
+
+        selfp.serdes.i = 0b01111111  # falling edge at fine_ts = 7
+        yield
+        selfp.serdes.i = 0b00000000
+        yield
+        self.check_input(selfp, stb=1, fine_ts=7)
+
+        selfp.serdes.i = 0b11000000  # rising edge at fine_ts = 6
+        yield
+        selfp.serdes.i = 0b11111111
+        yield
+        self.check_input(selfp, stb=1, fine_ts=6)
+
+        selfp.dut.rtlink.o.address = 2
+        selfp.dut.rtlink.o.data = 0b11
+        selfp.dut.rtlink.o.stb = 1  # set sensitivity to rising only
+        yield
+        selfp.dut.rtlink.o.stb = 0
+        yield
+
+        selfp.serdes.i = 0b00001111  # falling edge at fine_ts = 4
+        yield
+        self.check_input(selfp, stb=0)  # no strobe, sensitivity is rising edge
+
+        selfp.serdes.i = 0b11110000  # rising edge at fine_ts = 4
+        yield
+        self.check_input(selfp, stb=1, fine_ts=4)
+
+        selfp.dut.rtlink.o.address = 1
+        selfp.dut.rtlink.o.data = 1
+        selfp.dut.rtlink.o.stb = 1  # set Output Enable to 1
+        yield
+        selfp.dut.rtlink.o.stb = 0
+        yield
+        yield
+        self.check_output_enable(selfp, 1)
+
+        selfp.dut.rtlink.o.address = 0
+        selfp.dut.rtlink.o.data = 1
+        selfp.dut.rtlink.o.fine_ts = 3
+        selfp.dut.rtlink.o.stb = 1  # rising edge at fine_ts = 3
+        yield
+        selfp.dut.rtlink.o.stb = 0
+        yield
+        self.check_output(selfp, data=0b11111000)
+
+        yield
+        self.check_output(selfp, data=0xFF)  # stays at 1
+        
+        selfp.dut.rtlink.o.data = 0
+        selfp.dut.rtlink.o.fine_ts = 0
+        selfp.dut.rtlink.o.stb = 1  # falling edge at fine_ts = 0
+        yield
+        selfp.dut.rtlink.o.stb = 0
+        yield
+        self.check_output(selfp, data=0)
+
+        yield
+        self.check_output(selfp, data=0)
+        
+        selfp.dut.rtlink.o.data = 1
+        selfp.dut.rtlink.o.fine_ts = 7
+        selfp.dut.rtlink.o.stb = 1  # rising edge at fine_ts = 7
+        yield
+        selfp.dut.rtlink.o.stb = 0
+        yield
+        self.check_output(selfp, data=0b10000000)
+
+
+if __name__ == "__main__":
+    import sys
+    from migen.sim.generic import Simulator, TopLevel
+
+    if len(sys.argv) != 2:
+        print("Incorrect command line")
+        sys.exit(1)
+
+    cls = {
+        "output": _OutputTB,
+        "inout": _InoutTB
+    }[sys.argv[1]]
+
+    with Simulator(cls(), TopLevel("top.vcd", clk_period=int(1/0.125))) as s:
+        s.run()
diff --git a/artiq/gateware/rtio/phy/ttl_serdes_spartan6.py b/artiq/gateware/rtio/phy/ttl_serdes_spartan6.py
new file mode 100644
index 000000000..cf31449c1
--- /dev/null
+++ b/artiq/gateware/rtio/phy/ttl_serdes_spartan6.py
@@ -0,0 +1,154 @@
+from migen.fhdl.std import *
+
+from artiq.gateware.rtio.phy import ttl_serdes_generic
+
+
+class _OSERDES2_8X(Module):
+    def __init__(self, pad, stb):
+        self.o = Signal(8)
+        self.t_in = Signal()
+        self.t_out = Signal()
+
+        # # #
+
+        cascade = Signal(4)
+        o = self.o
+        common = dict(p_DATA_RATE_OQ="SDR", p_DATA_RATE_OT="SDR",
+                      p_DATA_WIDTH=8, p_OUTPUT_MODE="SINGLE_ENDED", i_TRAIN=0,
+                      i_CLK0=ClockSignal("rtiox8"), i_CLK1=0,
+                      i_CLKDIV=ClockSignal("rio_phy"),
+                      i_IOCE=stb, i_OCE=1, i_TCE=1, i_RST=0,
+                      i_T4=self.t_in, i_T3=self.t_in,
+                      i_T2=self.t_in, i_T1=self.t_in)
+
+        self.specials += [
+            Instance("OSERDES2", p_SERDES_MODE="MASTER",
+                     i_D4=o[7], i_D3=o[6], i_D2=o[5], i_D1=o[4],
+                     i_SHIFTIN1=1, i_SHIFTIN2=1,
+                     i_SHIFTIN3=cascade[2], i_SHIFTIN4=cascade[3],
+                     o_SHIFTOUT1=cascade[0], o_SHIFTOUT2=cascade[1],
+                     o_OQ=pad, o_TQ=self.t_out, **common),
+            Instance("OSERDES2", p_SERDES_MODE="SLAVE",
+                     i_D4=o[3], i_D3=o[2], i_D2=o[1], i_D1=o[0],
+                     i_SHIFTIN1=cascade[0], i_SHIFTIN2=cascade[1],
+                     i_SHIFTIN3=1, i_SHIFTIN4=1,
+                     o_SHIFTOUT3=cascade[2], o_SHIFTOUT4=cascade[3],
+                     **common),
+        ]
+
+
+class _IOSERDES2_8X(Module):
+    def __init__(self, pad, stb):
+        self.o = Signal(8)
+        self.i = Signal(8)
+        self.oe = Signal()
+
+        # # #
+
+        pad_i = Signal()
+        pad_o = Signal()
+        cascade = Signal()
+        i = self.i
+        common = dict(p_BITSLIP_ENABLE="FALSE", p_DATA_RATE="SDR",
+                      p_DATA_WIDTH=8, p_INTERFACE_TYPE="RETIMED",
+                      i_BITSLIP=0, i_CE0=1, i_IOCE=stb,
+                      i_RST=0, i_CLK0=ClockSignal("rtiox8"), i_CLK1=0,
+                      i_CLKDIV=ClockSignal("rio_phy"))
+        self.specials += [
+            Instance("ISERDES2", p_SERDES_MODE="MASTER",
+                     o_Q4=i[7], o_Q3=i[6], o_Q2=i[5], o_Q1=i[4],
+                     o_SHIFTOUT=cascade, i_D=pad_i, i_SHIFTIN=0,
+                     **common),
+            Instance("ISERDES2", p_SERDES_MODE="SLAVE",
+                     o_Q4=i[3], o_Q3=i[2], o_Q2=i[1], o_Q1=i[0],
+                     i_D=0, i_SHIFTIN=cascade, **common),
+        ]
+
+        oserdes = _OSERDES2_8X(pad_o, stb)
+        self.submodules += oserdes
+        self.specials += Instance("IOBUF",
+                                  i_I=pad_o, o_O=pad_i, i_T=oserdes.t_out,
+                                  io_IO=pad)
+        self.comb += [
+            oserdes.t_in.eq(~self.oe),
+            oserdes.o.eq(self.o),
+        ]
+
+
+class Output_8X(ttl_serdes_generic.Output):
+    def __init__(self, pad, stb):
+        serdes = _OSERDES2_8X(pad, stb)
+        self.submodules += serdes
+        ttl_serdes_generic.Output.__init__(self, serdes)
+
+
+class Inout_8X(ttl_serdes_generic.Inout):
+    def __init__(self, pad, stb):
+        serdes = _IOSERDES2_8X(pad, stb)
+        self.submodules += serdes
+        ttl_serdes_generic.Inout.__init__(self, serdes)
+
+
+class _OSERDES2_4X(Module):
+    def __init__(self, pad, stb):
+        self.o = Signal(4)
+        self.t_in = Signal()
+        self.t_out = Signal()
+
+        # # #
+
+        o = self.o
+        self.specials += Instance("OSERDES2", p_SERDES_MODE="NONE",
+                                  p_DATA_RATE_OQ="SDR", p_DATA_RATE_OT="SDR",
+                                  p_DATA_WIDTH=4, p_OUTPUT_MODE="SINGLE_ENDED",
+                                  i_TRAIN=0, i_CLK0=ClockSignal("rtiox4"),
+                                  i_CLK1=0, i_CLKDIV=ClockSignal("rio_phy"),
+                                  i_IOCE=stb, i_OCE=1, i_TCE=1, i_RST=0,
+                                  i_T4=self.t_in, i_T3=self.t_in,
+                                  i_T2=self.t_in, i_T1=self.t_in,
+                                  i_D4=o[3], i_D3=o[2], i_D2=o[1], i_D1=o[0],
+                                  o_OQ=pad, o_TQ=self.t_out)
+
+
+class _IOSERDES2_4X(Module):
+    def __init__(self, pad, stb):
+        self.o = Signal(4)
+        self.i = Signal(4)
+        self.oe = Signal()
+
+        # # #
+
+        pad_i = Signal()
+        pad_o = Signal()
+        i = self.i
+        self.specials += Instance("ISERDES2", p_SERDES_MODE="NONE",
+                                  p_BITSLIP_ENABLE="FALSE", p_DATA_RATE="SDR",
+                                  p_DATA_WIDTH=4, p_INTERFACE_TYPE="RETIMED",
+                                  i_BITSLIP=0, i_CE0=1, i_IOCE=stb,
+                                  i_RST=0, i_CLK0=ClockSignal("rtiox4"),
+                                  i_CLK1=0, i_CLKDIV=ClockSignal("rio_phy"),
+                                  o_Q4=i[3], o_Q3=i[2], o_Q2=i[1], o_Q1=i[0],
+                                  i_D=pad_i, i_SHIFTIN=0)
+        oserdes = _OSERDES2_4X(pad_o, stb)
+        self.submodules += oserdes
+        self.specials += Instance("IOBUF",
+                                  i_I=pad_o, o_O=pad_i, i_T=oserdes.t_out,
+                                  io_IO=pad)
+        self.comb += [
+            oserdes.t_in.eq(~self.oe),
+            oserdes.o.eq(self.o),
+        ]
+
+
+class Output_4X(ttl_serdes_generic.Output):
+    def __init__(self, pad, stb):
+        serdes = _OSERDES2_4X(pad, stb)
+        self.submodules += serdes
+        ttl_serdes_generic.Output.__init__(self, serdes)
+
+
+class Inout_4X(ttl_serdes_generic.Inout):
+    def __init__(self, pad, stb):
+        serdes = _IOSERDES2_4X(pad, stb)
+        self.submodules += serdes
+        ttl_serdes_generic.Inout.__init__(self, serdes)
diff --git a/artiq/test/coredevice.py b/artiq/test/coredevice.py
index 95e8573b9..e4d9375ba 100644
--- a/artiq/test/coredevice.py
+++ b/artiq/test/coredevice.py
@@ -120,7 +120,7 @@ class LoopbackCount(EnvExperiment):
     @kernel
     def run(self):
         self.ttl_inout.output()
-        delay(1*us)
+        delay(5*us)
         with parallel:
             self.ttl_inout.gate_rising(10*us)
             with sequential:
diff --git a/doc/manual/installing.rst b/doc/manual/installing.rst
index ff74515ba..5ade4cb7e 100644
--- a/doc/manual/installing.rst
+++ b/doc/manual/installing.rst
@@ -147,7 +147,7 @@ These steps are required to generate bitstream (``.bit``) files, build the MiSoC
     ::
 
         $ cd ~/artiq-dev
-        $ svn co https://xc3sprog.svn.sourceforge.net/svnroot/xc3sprog/trunk xc3sprog
+        $ svn co http://svn.code.sf.net/p/xc3sprog/code/trunk xc3sprog
         $ cd xc3sprog
         $ cmake . && make
         $ sudo make install
@@ -299,16 +299,14 @@ Installing the host-side software
 * Install the llvmlite Python bindings: ::
 
         $ cd ~/artiq-dev
-        $ git clone https://github.com/numba/llvmlite
+        $ git clone https://github.com/m-labs/llvmlite
+        $ git checkout backport-3.5
         $ cd llvmlite
         $ patch -p1 < ~/artiq-dev/artiq/misc/llvmlite-add-all-targets.patch
         $ patch -p1 < ~/artiq-dev/artiq/misc/llvmlite-rename.patch
         $ patch -p1 < ~/artiq-dev/artiq/misc/llvmlite-build-as-debug-on-windows.patch
         $ LLVM_CONFIG=/usr/local/llvm-or1k/bin/llvm-config python3 setup.py install --user
 
-.. note::
-    llvmlite is in development and its API is not stable yet. Commit ID ``11a8303d02e3d6dd2d1e0e9065701795cd8a979f`` is known to work.
-
 * Install ARTIQ: ::
 
         $ cd ~/artiq-dev
diff --git a/examples/master/ddb.pyon b/examples/master/ddb.pyon
index e853675c5..a1ce5680a 100644
--- a/examples/master/ddb.pyon
+++ b/examples/master/ddb.pyon
@@ -9,7 +9,7 @@
         "type": "local",
         "module": "artiq.coredevice.core",
         "class": "Core",
-        "arguments": {}
+        "arguments": {"ref_period": 1e-9}
     },
 
     "pmt0": {
@@ -133,6 +133,9 @@
         "command": "lda_controller -p {port} --bind {bind}"
     },
 
+    "ttl_inout": "pmt0",
+    "ttl_out": "ttl0",
+
     "pmt": "pmt0",
     "bd_dds": "dds0",
     "bd_sw": "ttl0",
diff --git a/examples/master/repository/tdr.py b/examples/master/repository/tdr.py
new file mode 100644
index 000000000..9a07d5753
--- /dev/null
+++ b/examples/master/repository/tdr.py
@@ -0,0 +1,71 @@
+from artiq import *
+
+
+class PulseNotReceivedError(Exception):
+    pass
+
+
+class TDR(EnvExperiment):
+    """Time domain reflectometer.
+
+    From ttl2 an impedance matched pulse is send onto a coax
+    cable with an open end. pmt0 (very short stub, high impedance) also
+    listens on the transmission line near ttl2.
+
+    When the forward propagating pulse passes pmt0, the voltage is half of the
+    logic voltage and does not register as a rising edge. Once the
+    rising edge is reflected at an open end (same sign) and passes by pmt0 on
+    its way back to ttl2, it is detected. Analogously, hysteresis leads to
+    detection of the falling edge once the reflection reaches pmt0 after
+    one round trip time.
+
+    This works marginally and is just a proof of principle: it relies on
+    hysteresis at FPGA inputs around half voltage and good impedance steps,
+    as well as reasonably low loss cable. It does not work well for longer
+    cables (>100 ns RTT). The default drive strength of 12 mA and 3.3 V would
+    be ~300 Ω but it seems 40 Ω series impedance at the output matches
+    the hysteresis of the input.
+
+    This is also equivalent to a loopback tester or a delay measurement.
+    """
+    def build(self):
+        self.attr_device("core")
+        self.attr_device("pmt0")
+        self.attr_device("ttl2")
+
+    def run(self):
+        n = 1000  # repetitions
+        latency = 50e-9  # calibrated latency without a transmission line
+        pulse = 1e-6  # pulse length, larger than rtt
+        try:
+            self.many(n, seconds_to_mu(pulse, self.core))
+        except PulseNotReceivedError:
+            print("to few edges: cable too long or wiring bad")
+        else:
+            print(self.t)
+            t_rise = mu_to_seconds(self.t[0], self.core)/n - latency
+            t_fall = mu_to_seconds(self.t[1], self.core)/n - latency - pulse
+            print("round trip times:")
+            print("rising: {:5g} ns, falling {:5g} ns".format(
+                t_rise/1e-9, t_fall/1e-9))
+
+    @kernel
+    def many(self, n, p):
+        t = [0 for i in range(2)]
+        self.core.break_realtime()
+        for i in range(n):
+            self.one(t, p)
+        self.t = t
+
+    @kernel
+    def one(self, t, p):
+        t0 = now_mu()
+        with parallel:
+            self.pmt0.gate_both_mu(2*p)
+            self.ttl2.pulse_mu(p)
+        for i in range(len(t)):
+            ti = self.pmt0.timestamp_mu()
+            if ti <= 0:
+                raise PulseNotReceivedError
+            t[i] += ti - t0
+        self.pmt0.count()  # flush
diff --git a/soc/runtime/Makefile b/soc/runtime/Makefile
index 241448f37..673235928 100644
--- a/soc/runtime/Makefile
+++ b/soc/runtime/Makefile
@@ -1,6 +1,6 @@
 include $(MSCDIR)/software/common.mak
 
-OBJECTS := isr.o flash_storage.o clock.o elf_loader.o services.o session.o log.o test_mode.o kloader.o bridge_ctl.o mailbox.o ksupport_data.o kserver.o moninj.o main.o
+OBJECTS := isr.o flash_storage.o clock.o rtiocrg.o elf_loader.o services.o session.o log.o test_mode.o kloader.o bridge_ctl.o mailbox.o ksupport_data.o kserver.o moninj.o main.o
 OBJECTS_KSUPPORT := ksupport.o exception_jmp.o exceptions.o mailbox.o bridge.o rtio.o ttl.o dds.o
 
 CFLAGS += -Ilwip/src/include -Iliblwip
diff --git a/soc/runtime/bridge.c b/soc/runtime/bridge.c
index eb16587de..78083a16f 100644
--- a/soc/runtime/bridge.c
+++ b/soc/runtime/bridge.c
@@ -5,12 +5,14 @@
 #include "dds.h"
 #include "bridge.h"
 
+#define TIME_BUFFER (8000 << RTIO_FINE_TS_WIDTH)
+
 static void dds_write(int addr, int data)
 {
     rtio_chan_sel_write(RTIO_DDS_CHANNEL);
     rtio_o_address_write(addr);
     rtio_o_data_write(data);
-    rtio_o_timestamp_write(rtio_get_counter() + 8000);
+    rtio_o_timestamp_write(rtio_get_counter() + TIME_BUFFER);
     rtio_o_we_write(1);
 }
 
@@ -46,7 +48,7 @@ void bridge_main(void)
                 struct msg_brg_ttl_out *msg;
 
                 msg = (struct msg_brg_ttl_out *)umsg;
-                ttl_set_oe(rtio_get_counter() + 8000, msg->channel, msg->value);
+                ttl_set_oe(rtio_get_counter() + TIME_BUFFER, msg->channel, msg->value);
                 mailbox_acknowledge();
                 break;
             }
@@ -54,7 +56,7 @@ void bridge_main(void)
                 struct msg_brg_ttl_out *msg;
 
                 msg = (struct msg_brg_ttl_out *)umsg;
-                ttl_set_o(rtio_get_counter() + 8000, msg->channel, msg->value);
+                ttl_set_o(rtio_get_counter() + TIME_BUFFER, msg->channel, msg->value);
                 mailbox_acknowledge();
                 break;
             }
diff --git a/soc/runtime/clock.c b/soc/runtime/clock.c
index 6b3946790..171fb6f30 100644
--- a/soc/runtime/clock.c
+++ b/soc/runtime/clock.c
@@ -26,6 +26,16 @@ long long int clock_get_ms(void)
     return clock_ms;
 }
 
+void busywait_us(long long int us)
+{
+    long long int threshold;
+
+    timer0_update_value_write(1);
+    threshold = timer0_value_read() - us*(long long int)identifier_frequency_read()/1000000LL;
+    while(timer0_value_read() > threshold)
+        timer0_update_value_write(1);
+}
+
 struct watchdog {
     int active;
     long long int threshold;
diff --git a/soc/runtime/clock.h b/soc/runtime/clock.h
index 70de5e5b7..4f4895443 100644
--- a/soc/runtime/clock.h
+++ b/soc/runtime/clock.h
@@ -3,6 +3,7 @@
 
 void clock_init(void);
 long long int clock_get_ms(void);
+void busywait_us(long long us);
 
 #define MAX_WATCHDOGS 16
 
diff --git a/soc/runtime/kloader.h b/soc/runtime/kloader.h
index 098a740b7..ceefbc89d 100644
--- a/soc/runtime/kloader.h
+++ b/soc/runtime/kloader.h
@@ -2,7 +2,7 @@
 #define __KLOADER_H
 
 #define KERNELCPU_EXEC_ADDRESS 0x40400000
-#define KERNELCPU_PAYLOAD_ADDRESS 0x40404000
+#define KERNELCPU_PAYLOAD_ADDRESS 0x40408000
 
 extern long long int now;
 
diff --git a/soc/runtime/ksupport.c b/soc/runtime/ksupport.c
index 032fe88ec..eb7a784b7 100644
--- a/soc/runtime/ksupport.c
+++ b/soc/runtime/ksupport.c
@@ -79,7 +79,7 @@ long long int now_init(void)
 
     if(now < 0) {
         rtio_init();
-        now = rtio_get_counter() + 125000;
+        now = rtio_get_counter() + (125000 << RTIO_FINE_TS_WIDTH);
     }
 
     return now;
diff --git a/soc/runtime/ksupport.ld b/soc/runtime/ksupport.ld
index 4126b4a7b..3cc585399 100644
--- a/soc/runtime/ksupport.ld
+++ b/soc/runtime/ksupport.ld
@@ -4,10 +4,10 @@ ENTRY(_start)
 INCLUDE generated/regions.ld
 
 /* First 4M of main memory are reserved for runtime code/data
- * then comes kernel memory. First 16K of kernel memory are for support code.
+ * then comes kernel memory. First 32K of kernel memory are for support code.
  */
 MEMORY {
-    ksupport : ORIGIN = 0x40400000, LENGTH = 0x4000
+    ksupport : ORIGIN = 0x40400000, LENGTH = 0x8000
 }
 
 /* On AMP systems, kernel stack is at the end of main RAM,
@@ -24,17 +24,6 @@ SECTIONS
         _etext = .;
     } > ksupport
 
-    .got :
-    {
-        _GLOBAL_OFFSET_TABLE_ = .;
-        *(.got)
-    } > ksupport
-
-    .got.plt :
-    {
-        *(.got.plt)
-    } > ksupport
-
     .rodata :
     {
         . = ALIGN(4);
diff --git a/soc/runtime/liblwip/Makefile b/soc/runtime/liblwip/Makefile
index 5dc22bcf3..0aaabd954 100644
--- a/soc/runtime/liblwip/Makefile
+++ b/soc/runtime/liblwip/Makefile
@@ -53,4 +53,3 @@ clean:
 
 liblwip.a: $(LWIPOBJS)
 	$(AR) clr liblwip.a $(LWIPOBJS)
-	$(RANLIB) liblwip.a
diff --git a/soc/runtime/linker.ld b/soc/runtime/linker.ld
index d0c3a6558..4a4217f1e 100644
--- a/soc/runtime/linker.ld
+++ b/soc/runtime/linker.ld
@@ -26,17 +26,6 @@ SECTIONS
 		_etext = .;
 	} > runtime
 
-	.got :
-	{
-		_GLOBAL_OFFSET_TABLE_ = .;
-		*(.got)
-	} > runtime
-
-	.got.plt :
-	{
-		*(.got.plt)
-	} > runtime
-
 	.rodata :
 	{
 		. = ALIGN(4);
diff --git a/soc/runtime/main.c b/soc/runtime/main.c
index 306f26b20..74df38aea 100644
--- a/soc/runtime/main.c
+++ b/soc/runtime/main.c
@@ -24,6 +24,7 @@
 #include "kloader.h"
 #include "flash_storage.h"
 #include "clock.h"
+#include "rtiocrg.h"
 #include "test_mode.h"
 #include "kserver.h"
 #include "session.h"
@@ -250,6 +251,7 @@ int main(void)
     puts("ARTIQ runtime built "__DATE__" "__TIME__"\n");
 
     clock_init();
+    rtiocrg_init();
     puts("Press 't' to enter test mode...");
     blink_led();
 
diff --git a/soc/runtime/rtiocrg.c b/soc/runtime/rtiocrg.c
new file mode 100644
index 000000000..255f2a308
--- /dev/null
+++ b/soc/runtime/rtiocrg.c
@@ -0,0 +1,67 @@
+#include <stdio.h>
+#include <generated/csr.h>
+
+#include "clock.h"
+#include "flash_storage.h"
+#include "rtiocrg.h"
+
+void rtiocrg_init(void)
+{
+    char b;
+    int clk;
+
+#ifdef CSR_RTIO_CRG_PLL_RESET_ADDR
+    rtio_crg_pll_reset_write(0);
+#endif
+    b = 'i';
+    clk = 0;
+    fs_read("startup_clock", &b, 1, NULL);
+    if(b == 'i')
+        printf("Startup RTIO clock: internal\n");
+    else if(b == 'e') {
+        printf("Startup RTIO clock: external\n");
+        clk = 1;
+    } else
+        printf("WARNING: unknown startup_clock entry in flash storage\n");
+
+    if(!rtiocrg_switch_clock(clk)) {
+        printf("WARNING: startup RTIO clock failed\n");
+        printf("WARNING: this may cause the system initialization to fail\n");
+        printf("WARNING: fix clocking and reset the device\n");
+    }
+}
+
+int rtiocrg_check(void)
+{
+#ifdef CSR_RTIO_CRG_PLL_RESET_ADDR
+    return rtio_crg_pll_locked_read();
+#else
+    return 1;
+#endif
+}
+
+int rtiocrg_switch_clock(int clk)
+{
+    int current_clk;
+
+    current_clk = rtio_crg_clock_sel_read();
+    if(clk == current_clk) {
+#ifdef CSR_RTIO_CRG_PLL_RESET_ADDR
+        busywait_us(150);
+        if(!rtio_crg_pll_locked_read())
+            return 0;
+#endif
+        return 1;
+    }
+#ifdef CSR_RTIO_CRG_PLL_RESET_ADDR
+    rtio_crg_pll_reset_write(1);
+#endif
+    rtio_crg_clock_sel_write(clk);
+#ifdef CSR_RTIO_CRG_PLL_RESET_ADDR
+    rtio_crg_pll_reset_write(0);
+    busywait_us(150);
+    if(!rtio_crg_pll_locked_read())
+        return 0;
+#endif
+    return 1;
+}
diff --git a/soc/runtime/rtiocrg.h b/soc/runtime/rtiocrg.h
new file mode 100644
index 000000000..df498f617
--- /dev/null
+++ b/soc/runtime/rtiocrg.h
@@ -0,0 +1,8 @@
+#ifndef __RTIOCRG_H
+#define __RTIOCRG_H
+
+void rtiocrg_init(void);
+int rtiocrg_check(void);
+int rtiocrg_switch_clock(int clk);
+
+#endif /* __RTIOCRG_H */
diff --git a/soc/runtime/session.c b/soc/runtime/session.c
index 9b3591e30..c0908639c 100644
--- a/soc/runtime/session.c
+++ b/soc/runtime/session.c
@@ -12,6 +12,7 @@
 #include "kloader.h"
 #include "exceptions.h"
 #include "flash_storage.h"
+#include "rtiocrg.h"
 #include "session.h"
 
 #define BUFFER_IN_SIZE (1024*1024)
@@ -154,8 +155,10 @@ static int process_input(void)
                 submit_output(9);
                 break;
             }
-            rtio_crg_clock_sel_write(buffer_in[9]);
-            buffer_out[8] = REMOTEMSG_TYPE_CLOCK_SWITCH_COMPLETED;
+            if(rtiocrg_switch_clock(buffer_in[9]))
+                buffer_out[8] = REMOTEMSG_TYPE_CLOCK_SWITCH_COMPLETED;
+            else
+                buffer_out[8] = REMOTEMSG_TYPE_CLOCK_SWITCH_FAILED;
             submit_output(9);
             break;
         case REMOTEMSG_TYPE_LOAD_OBJECT:
@@ -527,10 +530,17 @@ void session_poll(void **data, int *len)
 {
     int l;
 
-    if((user_kernel_state == USER_KERNEL_RUNNING) && watchdog_expired()) {
-        log("Watchdog expired");
-        *len = -1;
-        return;
+    if(user_kernel_state == USER_KERNEL_RUNNING) {
+        if(watchdog_expired()) {
+            log("Watchdog expired");
+            *len = -1;
+            return;
+        }
+        if(!rtiocrg_check()) {
+            log("RTIO clock failure");
+            *len = -1;
+            return;
+        }
     }
 
     l = get_out_packet_len();
diff --git a/soc/targets/artiq_kc705.py b/soc/targets/artiq_kc705.py
index cfc9774f9..a768e2f07 100644
--- a/soc/targets/artiq_kc705.py
+++ b/soc/targets/artiq_kc705.py
@@ -1,8 +1,11 @@
 from migen.fhdl.std import *
+from migen.genlib.resetsync import AsyncResetSynchronizer
+from migen.genlib.cdc import MultiReg
 from migen.bank.description import *
 from migen.bank import wbgen
 from mibuild.generic_platform import *
 from mibuild.xilinx.vivado import XilinxVivadoToolchain
+from mibuild.xilinx.ise import XilinxISEToolchain
 
 from misoclib.com import gpio
 from misoclib.soc import mem_decoder
@@ -11,13 +14,22 @@ from targets.kc705 import MiniSoC
 
 from artiq.gateware.soc import AMPSoC
 from artiq.gateware import rtio, nist_qc1, nist_qc2
-from artiq.gateware.rtio.phy import ttl_simple, dds
+from artiq.gateware.rtio.phy import ttl_simple, ttl_serdes_7series, dds
 
 
 class _RTIOCRG(Module, AutoCSR):
     def __init__(self, platform, rtio_internal_clk):
         self._clock_sel = CSRStorage()
-        self.clock_domains.cd_rtio = ClockDomain(reset_less=True)
+        self._pll_reset = CSRStorage(reset=1)
+        self._pll_locked = CSRStatus()
+        self.clock_domains.cd_rtio = ClockDomain()
+        self.clock_domains.cd_rtiox4 = ClockDomain(reset_less=True)
+
+        # 10 MHz when using 125MHz input
+        self.clock_domains.cd_ext_clkout = ClockDomain(reset_less=True)
+        ext_clkout = platform.request("user_sma_gpio_p")
+        self.sync.ext_clkout += ext_clkout.eq(~ext_clkout)
+
 
         rtio_external_clk = Signal()
         user_sma_clock = platform.request("user_sma_clock")
@@ -25,11 +37,40 @@ class _RTIOCRG(Module, AutoCSR):
         self.specials += Instance("IBUFDS",
                                   i_I=user_sma_clock.p, i_IB=user_sma_clock.n,
                                   o_O=rtio_external_clk)
-        self.specials += Instance("BUFGMUX",
-                                  i_I0=rtio_internal_clk,
-                                  i_I1=rtio_external_clk,
-                                  i_S=self._clock_sel.storage,
-                                  o_O=self.cd_rtio.clk)
+
+        pll_locked = Signal()
+        rtio_clk = Signal()
+        rtiox4_clk = Signal()
+        ext_clkout_clk = Signal()
+        self.specials += [
+            Instance("PLLE2_ADV",
+                     p_STARTUP_WAIT="FALSE", o_LOCKED=pll_locked,
+
+                     p_REF_JITTER1=0.01,
+                     p_CLKIN1_PERIOD=8.0, p_CLKIN2_PERIOD=8.0,
+                     i_CLKIN1=rtio_internal_clk, i_CLKIN2=rtio_external_clk,
+                     # Warning: CLKINSEL=0 means CLKIN2 is selected
+                     i_CLKINSEL=~self._clock_sel.storage,
+
+                     # VCO @ 1GHz when using 125MHz input
+                     p_CLKFBOUT_MULT=8, p_DIVCLK_DIVIDE=1,
+                     i_CLKFBIN=self.cd_rtio.clk,
+                     i_RST=self._pll_reset.storage,
+
+                     o_CLKFBOUT=rtio_clk,
+
+                     p_CLKOUT0_DIVIDE=2, p_CLKOUT0_PHASE=0.0,
+                     o_CLKOUT0=rtiox4_clk,
+
+                     p_CLKOUT1_DIVIDE=50, p_CLKOUT1_PHASE=0.0,
+                     o_CLKOUT1=ext_clkout_clk),
+            Instance("BUFG", i_I=rtio_clk, o_O=self.cd_rtio.clk),
+            Instance("BUFG", i_I=rtiox4_clk, o_O=self.cd_rtiox4.clk),
+            Instance("BUFG", i_I=ext_clkout_clk, o_O=self.cd_ext_clkout.clk),
+
+            AsyncResetSynchronizer(self.cd_rtio, ~pll_locked),
+            MultiReg(pll_locked, self._pll_locked.status)
+        ]
 
 
 class _NIST_QCx(MiniSoC, AMPSoC):
@@ -57,10 +98,10 @@ class _NIST_QCx(MiniSoC, AMPSoC):
             platform.request("user_led", 1)))
 
     def add_rtio(self, rtio_channels):
-        self.submodules.rtio_crg = _RTIOCRG(self.platform, self.crg.pll_sys)
-        self.submodules.rtio = rtio.RTIO(rtio_channels,
-                                         clk_freq=125000000)
+        self.submodules.rtio_crg = _RTIOCRG(self.platform, self.crg.cd_sys.clk)
+        self.submodules.rtio = rtio.RTIO(rtio_channels)
         self.add_constant("RTIO_FINE_TS_WIDTH", self.rtio.fine_ts_width)
+        assert self.rtio.fine_ts_width <= 3
         self.add_constant("DDS_RTIO_CLK_RATIO", 8 >> self.rtio.fine_ts_width)
         self.submodules.rtio_moninj = rtio.MonInj(rtio_channels)
 
@@ -71,6 +112,13 @@ create_clock -name rio_clk -period 8.0 [get_nets {rio_clk}]
 set_false_path -from [get_clocks rsys_clk] -to [get_clocks rio_clk]
 set_false_path -from [get_clocks rio_clk] -to [get_clocks rsys_clk]
 """, rsys_clk=self.rtio.cd_rsys.clk, rio_clk=self.rtio.cd_rio.clk)
+        if isinstance(self.platform.toolchain, XilinxISEToolchain):
+            self.platform.add_platform_command("""
+NET "sys_clk" TNM_NET = "GRPrsys_clk";
+NET "{rio_clk}" TNM_NET = "GRPrio_clk";
+TIMESPEC "TSfix_cdc1" = FROM "GRPrsys_clk" TO "GRPrio_clk" TIG;
+TIMESPEC "TSfix_cdc2" = FROM "GRPrio_clk" TO "GRPrsys_clk" TIG;
+""", rio_clk=self.rtio_crg.cd_rtio.clk)
 
         rtio_csrs = self.rtio.get_csrs()
         self.submodules.rtiowb = wbgen.Bank(rtio_csrs)
@@ -92,11 +140,11 @@ class NIST_QC1(_NIST_QCx):
 
         rtio_channels = []
         for i in range(2):
-            phy = ttl_simple.Inout(platform.request("pmt", i))
+            phy = ttl_serdes_7series.Inout_8X(platform.request("pmt", i))
             self.submodules += phy
             rtio_channels.append(rtio.Channel.from_phy(phy, ififo_depth=512))
         for i in range(15):
-            phy = ttl_simple.Output(platform.request("ttl", i))
+            phy = ttl_serdes_7series.Output_8X(platform.request("ttl", i))
             self.submodules += phy
             rtio_channels.append(rtio.Channel.from_phy(phy))
 
@@ -131,11 +179,11 @@ class NIST_QC2(_NIST_QCx):
                 # TTL14 is for the clock generator
                 continue
             if i % 4 == 3:
-                phy = ttl_simple.Inout(platform.request("ttl", i))
+                phy = ttl_serdes_7series.Inout_8X(platform.request("ttl", i))
                 self.submodules += phy
                 rtio_channels.append(rtio.Channel.from_phy(phy, ififo_depth=512))
             else:
-                phy = ttl_simple.Output(platform.request("ttl", i))
+                phy = ttl_serdes_7series.Output_8X(platform.request("ttl", i))
                 self.submodules += phy
                 rtio_channels.append(rtio.Channel.from_phy(phy))
 
diff --git a/soc/targets/artiq_pipistrello.py b/soc/targets/artiq_pipistrello.py
index 23c50bc3d..7a16cdcfe 100644
--- a/soc/targets/artiq_pipistrello.py
+++ b/soc/targets/artiq_pipistrello.py
@@ -3,6 +3,8 @@ from fractions import Fraction
 from migen.fhdl.std import *
 from migen.bank.description import *
 from migen.bank import wbgen
+from migen.genlib.resetsync import AsyncResetSynchronizer
+from migen.genlib.cdc import MultiReg
 
 from misoclib.com import gpio
 from misoclib.soc import mem_decoder
@@ -11,46 +13,84 @@ from targets.pipistrello import BaseSoC
 
 from artiq.gateware.soc import AMPSoC
 from artiq.gateware import rtio, nist_qc1
-from artiq.gateware.rtio.phy import ttl_simple, dds
+from artiq.gateware.rtio.phy import ttl_simple, ttl_serdes_spartan6, dds
 
 
 class _RTIOCRG(Module, AutoCSR):
     def __init__(self, platform, clk_freq):
         self._clock_sel = CSRStorage()
-        self.clock_domains.cd_rtio = ClockDomain(reset_less=True)
+        self._pll_reset = CSRStorage(reset=1)
+        self._pll_locked = CSRStatus()
 
-        f = Fraction(125*1000*1000, clk_freq)
+        self.clock_domains.cd_rtio = ClockDomain()
+        self.clock_domains.cd_rtiox4 = ClockDomain(reset_less=True)
+        self.clock_domains.cd_rtiox8 = ClockDomain(reset_less=True)
+        self.rtiox4_stb = Signal()
+        self.rtiox8_stb = Signal()
+
+        rtio_f = 125*1000*1000
+        f = Fraction(rtio_f, clk_freq)
         rtio_internal_clk = Signal()
-        self.specials += Instance("DCM_CLKGEN",
-                                  p_CLKFXDV_DIVIDE=2,
-                                  p_CLKFX_DIVIDE=f.denominator,
-                                  p_CLKFX_MD_MAX=float(f),
-                                  p_CLKFX_MULTIPLY=f.numerator,
-                                  p_CLKIN_PERIOD=1e9/clk_freq,
-                                  p_SPREAD_SPECTRUM="NONE",
-                                  p_STARTUP_WAIT="FALSE",
-                                  i_CLKIN=ClockSignal(),
-                                  o_CLKFX=rtio_internal_clk,
-                                  i_FREEZEDCM=0,
-                                  i_RST=ResetSignal())
-
-        rtio_external_clk = platform.request("pmt", 2)
-        # ISE infers constraints for the internal clock
-        # and propagates them through the BUFGMUX. Adding this:
-        # platform.add_period_constraint(rtio_external_clk, 8.0)
-        # seems to confuse it
-        self.specials += Instance("BUFGMUX",
-                                  i_I0=rtio_internal_clk,
-                                  i_I1=rtio_external_clk,
-                                  i_S=self._clock_sel.storage,
-                                  o_O=self.cd_rtio.clk)
+        rtio_external_clk = Signal()
+        pmt2 = platform.request("pmt", 2)
+        dcm_locked = Signal()
+        rtio_clk = Signal()
+        pll_locked = Signal()
+        pll = Signal(3)
+        pll_fb = Signal()
+        self.specials += [
+            Instance("IBUFG", i_I=pmt2, o_O=rtio_external_clk),
+            Instance("DCM_CLKGEN", p_CLKFXDV_DIVIDE=2,
+                     p_CLKFX_DIVIDE=f.denominator, p_CLKFX_MD_MAX=float(f),
+                     p_CLKFX_MULTIPLY=f.numerator, p_CLKIN_PERIOD=1e9/clk_freq,
+                     p_SPREAD_SPECTRUM="NONE", p_STARTUP_WAIT="FALSE",
+                     i_CLKIN=ClockSignal(), o_CLKFX=rtio_internal_clk,
+                     i_FREEZEDCM=0, i_RST=ResetSignal(), o_LOCKED=dcm_locked),
+            Instance("BUFGMUX",
+                     i_I0=rtio_internal_clk, i_I1=rtio_external_clk,
+                     i_S=self._clock_sel.storage, o_O=rtio_clk),
+            Instance("PLL_ADV", p_SIM_DEVICE="SPARTAN6",
+                     p_BANDWIDTH="OPTIMIZED", p_COMPENSATION="INTERNAL",
+                     p_REF_JITTER=.01, p_CLK_FEEDBACK="CLKFBOUT",
+                     i_DADDR=0, i_DCLK=0, i_DEN=0, i_DI=0, i_DWE=0,
+                     i_RST=self._pll_reset.storage | ~dcm_locked, i_REL=0,
+                     p_DIVCLK_DIVIDE=1, p_CLKFBOUT_MULT=8,
+                     p_CLKFBOUT_PHASE=0., i_CLKINSEL=1,
+                     i_CLKIN1=rtio_clk, i_CLKIN2=0,
+                     p_CLKIN1_PERIOD=1e9/rtio_f, p_CLKIN2_PERIOD=0.,
+                     i_CLKFBIN=pll_fb, o_CLKFBOUT=pll_fb, o_LOCKED=pll_locked,
+                     o_CLKOUT0=pll[0], p_CLKOUT0_DUTY_CYCLE=.5,
+                     o_CLKOUT1=pll[1], p_CLKOUT1_DUTY_CYCLE=.5,
+                     o_CLKOUT2=pll[2], p_CLKOUT2_DUTY_CYCLE=.5,
+                     p_CLKOUT0_PHASE=0., p_CLKOUT0_DIVIDE=1,
+                     p_CLKOUT1_PHASE=0., p_CLKOUT1_DIVIDE=2,
+                     p_CLKOUT2_PHASE=0., p_CLKOUT2_DIVIDE=8),
+            Instance("BUFPLL", p_DIVIDE=8,
+                     i_PLLIN=pll[0], i_GCLK=self.cd_rtio.clk,
+                     i_LOCKED=pll_locked, o_IOCLK=self.cd_rtiox8.clk,
+                     o_SERDESSTROBE=self.rtiox8_stb),
+            Instance("BUFPLL", p_DIVIDE=4,
+                     i_PLLIN=pll[1], i_GCLK=self.cd_rtio.clk,
+                     i_LOCKED=pll_locked, o_IOCLK=self.cd_rtiox4.clk,
+                     o_SERDESSTROBE=self.rtiox4_stb),
+            Instance("BUFG", i_I=pll[2], o_O=self.cd_rtio.clk),
+            AsyncResetSynchronizer(self.cd_rtio, ~pll_locked),
+            MultiReg(pll_locked, self._pll_locked.status),
+        ]
 
+        # ISE infers correct period constraints for cd_rtio.clk from
+        # the internal clock. The first two TIGs target just the BUFGMUX.
         platform.add_platform_command("""
-NET "{int_clk}" TNM_NET = "GRPint_clk";
 NET "sys_clk" TNM_NET = "GRPsys_clk";
-TIMESPEC "TSfix_ise1" = FROM "GRPint_clk" TO "GRPsys_clk" TIG;
+NET "{ext_clk}" TNM_NET = "GRPext_clk";
+TIMESPEC "TSfix_ise1" = FROM "GRPsys_clk" TO "GRPext_clk" TIG;
+NET "{int_clk}" TNM_NET = "GRPint_clk";
 TIMESPEC "TSfix_ise2" = FROM "GRPsys_clk" TO "GRPint_clk" TIG;
-""", int_clk=rtio_internal_clk)
+NET "{rtio_clk}" TNM_NET = "GRPrtio_clk";
+TIMESPEC "TSfix_ise3" = FROM "GRPrtio_clk" TO "GRPsys_clk" TIG;
+TIMESPEC "TSfix_ise4" = FROM "GRPsys_clk" TO "GRPrtio_clk" TIG;
+""", ext_clk=rtio_external_clk, int_clk=rtio_internal_clk,
+     rtio_clk=self.cd_rtio.clk)
 
 
 class NIST_QC1(BaseSoC, AMPSoC):
@@ -73,6 +113,7 @@ class NIST_QC1(BaseSoC, AMPSoC):
                          sdram_controller_settings=MiniconSettings(l2_size=64*1024),
                          with_timer=False, **kwargs)
         AMPSoC.__init__(self)
+        platform.toolchain.bitgen_opt = "-g Binary:Yes -w"
         platform.toolchain.ise_commands += """
 trce -v 12 -fastpaths -tsi {build_name}.tsi -o {build_name}.twr {build_name}.ncd {build_name}.pcf
 """
@@ -90,16 +131,29 @@ trce -v 12 -fastpaths -tsi {build_name}.tsi -o {build_name}.twr {build_name}.ncd
             platform.request("ttl_h_tx_en").eq(1)
         ]
 
+        self.submodules.rtio_crg = _RTIOCRG(platform, self.clk_freq)
+
         # RTIO channels
         rtio_channels = []
+        # pmt1 can run on a 8x serdes if pmt0 is not used
         for i in range(2):
-            phy = ttl_simple.Inout(platform.request("pmt", i))
+            phy = ttl_serdes_spartan6.Inout_4X(platform.request("pmt", i),
+                                               self.rtio_crg.rtiox4_stb)
             self.submodules += phy
             rtio_channels.append(rtio.Channel.from_phy(phy, ififo_depth=512,
                                                        ofifo_depth=4))
 
+        # ttl2 can run on a 8x serdes if xtrig is not used
         for i in range(15):
-            phy = ttl_simple.Output(platform.request("ttl", i))
+            if i in (0, 1):
+                phy = ttl_serdes_spartan6.Output_4X(platform.request("ttl", i),
+                                                    self.rtio_crg.rtiox4_stb)
+            elif i in (2,):
+                phy = ttl_serdes_spartan6.Output_8X(platform.request("ttl", i),
+                                                    self.rtio_crg.rtiox8_stb)
+            else:
+                phy = ttl_simple.Output(platform.request("ttl", i))
+
             self.submodules += phy
             rtio_channels.append(rtio.Channel.from_phy(phy, ofifo_depth=256))
 
@@ -120,16 +174,16 @@ trce -v 12 -fastpaths -tsi {build_name}.tsi -o {build_name}.twr {build_name}.ncd
         self.add_constant("RTIO_DDS_CHANNEL", len(rtio_channels))
         self.add_constant("DDS_CHANNEL_COUNT", 8)
         self.add_constant("DDS_AD9858")
-        phy = dds.AD9858(platform.request("dds"), 8)
+        dds_pins = platform.request("dds")
+        self.comb += dds_pins.p.eq(0)
+        phy = dds.AD9858(dds_pins, 8)
         self.submodules += phy
         rtio_channels.append(rtio.Channel.from_phy(phy,
                                                    ofifo_depth=512,
                                                    ififo_depth=4))
 
         # RTIO core
-        self.submodules.rtio_crg = _RTIOCRG(platform, self.clk_freq)
-        self.submodules.rtio = rtio.RTIO(rtio_channels,
-                                         clk_freq=125000000)
+        self.submodules.rtio = rtio.RTIO(rtio_channels)
         self.add_constant("RTIO_FINE_TS_WIDTH", self.rtio.fine_ts_width)
         self.add_constant("DDS_RTIO_CLK_RATIO", 8 >> self.rtio.fine_ts_width)
         self.submodules.rtio_moninj = rtio.MonInj(rtio_channels)