diff --git a/neural_modelling/makefiles/local_only_combined/IF_curr_delta_conv/Makefile b/neural_modelling/makefiles/local_only_combined/IF_curr_delta_conv/Makefile
new file mode 100644
index 00000000000..5026a0cca88
--- /dev/null
+++ b/neural_modelling/makefiles/local_only_combined/IF_curr_delta_conv/Makefile
@@ -0,0 +1,26 @@
+# Copyright (c) 2021-2022 The University of Manchester
+#
+# This program is free software: you can redistribute it and/or modify
+# it under the terms of the GNU General Public License as published by
+# the Free Software Foundation, either version 3 of the License, or
+# (at your option) any later version.
+#
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU General Public License
+# along with this program.  If not, see <http://www.gnu.org/licenses/>.
+
+APP = $(notdir $(CURDIR))
+
+NEURON_MODEL = $(NEURON_DIR)/neuron/models/neuron_model_lif_impl.c
+NEURON_MODEL_H = $(NEURON_DIR)/neuron/models/neuron_model_lif_impl.h
+INPUT_TYPE_H = $(NEURON_DIR)/neuron/input_types/input_type_delta.h
+NEURON_IMPL_H = $(NEURON_DIR)/neuron/implementations/neuron_impl_standard.h
+THRESHOLD_TYPE_H = $(NEURON_DIR)/neuron/threshold_types/threshold_type_static.h
+SYNAPSE_TYPE_H = $(NEURON_DIR)/neuron/synapse_types/synapse_types_delta_impl.h
+LOCAL_ONLY_IMPL = $(NEURON_DIR)/neuron/local_only/local_only_conv_impl.c
+
+include ../local_only.mk
diff --git a/neural_modelling/makefiles/local_only_combined/Makefile b/neural_modelling/makefiles/local_only_combined/Makefile
index 4b67130b99a..b3b9193129b 100644
--- a/neural_modelling/makefiles/local_only_combined/Makefile
+++ b/neural_modelling/makefiles/local_only_combined/Makefile
@@ -13,7 +13,8 @@
 # You should have received a copy of the GNU General Public License
 # along with this program.  If not, see <http://www.gnu.org/licenses/>.
 
-MODELS = IF_curr_exp_conv\
+MODELS = IF_curr_delta_conv\
+		 IF_curr_exp_conv\
          IF_curr_exp_pool_dense
 
 all:
diff --git a/neural_modelling/src/neuron/local_only/local_only_conv_impl.c b/neural_modelling/src/neuron/local_only/local_only_conv_impl.c
index 5b82bdbd958..0d50f3e6848 100644
--- a/neural_modelling/src/neuron/local_only/local_only_conv_impl.c
+++ b/neural_modelling/src/neuron/local_only/local_only_conv_impl.c
@@ -60,9 +60,12 @@ typedef struct {
     lc_shape_t kernel;
     lc_shape_t padding;
     lc_coord_t recip_strides;
+    lc_coord_t strides;
     lc_coord_t recip_pool_strides;
     uint16_t positive_synapse_type;
     uint16_t negative_synapse_type;
+    uint32_t delay;
+    uint32_t strides_delay_step;
     lc_weight_t weights[]; // n_weights = next_even(kernel.width * kernel.height)
 } connector;
 
@@ -138,7 +141,15 @@ bool local_only_impl_initialise(void *address){
     return true;
 }
 
-//! \brief Multiply an integer by a 16-bit reciprocal and return the floored
+//! \brief Calculate the remainder from a division
+static inline int16_t calc_remainder(int16_t dividend, int16_t divisor, int16_t quotient) {
+    int16_t remainder = dividend - quotient * divisor;
+    log_debug("remainder: %d = %d * %d + %d",
+            dividend, quotient, divisor, remainder);
+    return remainder;
+}
+
+//! \brief Calculate remainder Multiply an integer by a 16-bit reciprocal and return the floored
 //!        integer result
 static inline int16_t recip_multiply(int16_t integer, int16_t recip) {
     int32_t i = integer;
@@ -146,19 +157,17 @@ static inline int16_t recip_multiply(int16_t integer, int16_t recip) {
     return (int16_t) ((i * r) >> RECIP_FRACT_BITS);
 }
 
-//! \brief Do a mapping from pre to post 2D spaces, we use the standard
-//! padding, kernel, strides from Convolutional Neural Networks
-//! because of the way we're looping through the kernel, we divide the kernel
-//! shape by 2.
-static inline lc_coord_t map_pre_to_post(connector *connector, lc_coord_t pre,
-        int16_t half_kh, int16_t half_kw) {
-    lc_coord_t post = pre;
-    post.col = recip_multiply(post.col, connector->recip_pool_strides.col);
-    post.row = recip_multiply(post.row, connector->recip_pool_strides.row);
-    post.col = post.col - half_kw + connector->padding.width;
-    post.row = post.row - half_kh + connector->padding.height;
-    post.col = recip_multiply(post.col, connector->recip_strides.col);
-    post.row = recip_multiply(post.row, connector->recip_strides.row);
+//! \brief Do a mapping from pre to post 2D spaces
+static inline lc_coord_t map_pre_to_post(connector *connector, lc_coord_t pre, lc_coord_t *start_i) {
+    pre.col = recip_multiply(pre.col, connector->recip_pool_strides.col);
+    pre.row = recip_multiply(pre.row, connector->recip_pool_strides.row);
+    pre.col += connector->padding.width;
+    pre.row += connector->padding.height;
+    lc_coord_t post;
+    post.col = recip_multiply(pre.col, connector->recip_strides.col);
+    post.row = recip_multiply(pre.row, connector->recip_strides.row);
+    start_i->col = calc_remainder(pre.col, connector->strides.col, post.col);
+    start_i->row = calc_remainder(pre.row, connector->strides.row, post.row);
     return post;
 }
 
@@ -169,21 +178,34 @@ static inline lc_coord_t map_pre_to_post(connector *connector, lc_coord_t pre,
 static inline void do_convolution_operation(
         uint32_t time, lc_coord_t pre_coord, connector *connector,
         uint16_t *ring_buffers) {
-    int32_t half_kh = connector->kernel.height / 2;
-    int32_t half_kw = connector->kernel.width / 2;
-    lc_coord_t post_coord = map_pre_to_post(connector, pre_coord, half_kh, half_kw);
+    lc_coord_t start_i;
+    log_debug("kernel height: %d, kernel width: %d, padding height: %d, padding width: %d, strides row: %d, strides col: %d", connector->kernel.height, connector->kernel.width, connector->padding.height, connector->padding.width, connector->strides.row, connector->strides.col);
+    lc_coord_t post_coord = map_pre_to_post(connector, pre_coord, &start_i);
     log_debug("pre row %d, col %d AS post row %d, col %d",
             pre_coord.row, pre_coord.col, post_coord.row, post_coord.col);
 
     int32_t kw = connector->kernel.width;
-    for (int32_t r = -half_kh, kr = 0; r <= half_kh; r++, kr++) {
-        int32_t tmp_row = post_coord.row + r;
+    for (int32_t i_row = start_i.row, tmp_row = post_coord.row; i_row < connector->kernel.height; i_row += connector->strides.row, --tmp_row) {
+        int32_t kr = connector->kernel.height - 1 - i_row;
+        log_debug("i_row = %u, kr = %u, tmp_row = %u", i_row, kr, tmp_row);
+
         if ((tmp_row < config.post_start.row) || (tmp_row > config.post_end.row)) {
+            log_debug("tmp_row outside");
             continue;
         }
-        for (int32_t c = -half_kw, kc = 0; c <= half_kw; c++, kc++) {
-            int32_t tmp_col = post_coord.col + c;
+
+        uint32_t delay = connector->delay;
+        if (connector->strides_delay_step != 0)
+        {
+            delay -= start_i.col * connector->strides_delay_step;
+            log_debug("start_i.col = %u, delay = %u", start_i.col, delay);
+        }
+        
+        for (int32_t i_col = start_i.col, tmp_col = post_coord.col; i_col < connector->kernel.width; i_col += connector->strides.col, --tmp_col) {
+            int32_t kc = connector->kernel.width - 1 - i_col;
+            log_debug("i_col = %u, kc = %u, tmp_col = %u", i_col, kc, tmp_col);
             if ((tmp_col < config.post_start.col) || (tmp_col > config.post_end.col)) {
+                log_debug("tmp_col outside");
                 continue;
             }
 
@@ -192,18 +214,20 @@ static inline void do_convolution_operation(
                 ((tmp_row - config.post_start.row) * config.post_shape.width)
                     + (tmp_col - config.post_start.col);
             uint32_t k = (kr * kw) + kc;
+            log_debug("weight index = %u", k);
             lc_weight_t weight = connector->weights[k];
             if (weight == 0) {
+                log_debug("zero weight");
                 continue;
             }
             uint32_t rb_index = 0;
             if (weight > 0) {
-                rb_index = synapse_row_get_ring_buffer_index(time + 1,
+                rb_index = synapse_row_get_ring_buffer_index(time + delay,
                     connector->positive_synapse_type, post_index,
                     synapse_type_index_bits, synapse_index_bits,
                     synapse_delay_mask);
             } else {
-                rb_index = synapse_row_get_ring_buffer_index(time + 1,
+                rb_index = synapse_row_get_ring_buffer_index(time + delay,
                     connector->negative_synapse_type, post_index,
                     synapse_type_index_bits, synapse_index_bits,
                     synapse_delay_mask);
diff --git a/neural_modelling/src/neuron/local_only/local_only_pool_dense_impl.c b/neural_modelling/src/neuron/local_only/local_only_pool_dense_impl.c
index 9dfbf3840c0..515c5bed3f5 100644
--- a/neural_modelling/src/neuron/local_only/local_only_pool_dense_impl.c
+++ b/neural_modelling/src/neuron/local_only/local_only_pool_dense_impl.c
@@ -68,6 +68,7 @@ typedef struct {
     uint32_t n_weights;
     uint16_t positive_synapse_type;
     uint16_t negative_synapse_type;
+    uint32_t delay;
     dimension dimensions[];
     // Also follows:
     // lc_weight_t weights[];
@@ -213,12 +214,12 @@ void local_only_impl_process_spike(
         }
         uint32_t rb_index = 0;
         if (weight > 0) {
-            rb_index = synapse_row_get_ring_buffer_index(time + 1,
+            rb_index = synapse_row_get_ring_buffer_index(time + connector->delay,
                 connector->positive_synapse_type, post_index,
                 synapse_type_index_bits, synapse_index_bits,
                 synapse_delay_mask);
         } else {
-            rb_index = synapse_row_get_ring_buffer_index(time + 1,
+            rb_index = synapse_row_get_ring_buffer_index(time + connector->delay,
                 connector->negative_synapse_type, post_index,
                 synapse_type_index_bits, synapse_index_bits,
                 synapse_delay_mask);
diff --git a/spynnaker/pyNN/models/neural_projections/connectors/convolution_connector.py b/spynnaker/pyNN/models/neural_projections/connectors/convolution_connector.py
index 55d699f5548..5dcf453bb93 100644
--- a/spynnaker/pyNN/models/neural_projections/connectors/convolution_connector.py
+++ b/spynnaker/pyNN/models/neural_projections/connectors/convolution_connector.py
@@ -27,13 +27,14 @@
 from spynnaker.pyNN.utilities.utility_calls import get_n_bits
 from spynnaker.pyNN.models.abstract_models import HasShapeKeyFields
 from spynnaker.pyNN.utilities.constants import SPIKE_PARTITION_ID
+from spynnaker.pyNN.data.spynnaker_data_view import SpynnakerDataView
 
 #: The number of 32-bit words in the source_key_info struct
 SOURCE_KEY_INFO_WORDS = 7
 
 #: The number of 16-bit shorts in the connector struct,
 #: ignoring the source_key_info struct and the weights (which are dynamic)
-CONNECTOR_CONFIG_SHORTS = 12
+CONNECTOR_CONFIG_SHORTS = 18
 
 
 class ConvolutionConnector(AbstractConnector):
@@ -50,13 +51,16 @@ class ConvolutionConnector(AbstractConnector):
         "__pool_shape",
         "__pool_stride",
         "__positive_receptor_type",
-        "__negative_receptor_type"
+        "__negative_receptor_type",
+        "__horizontal_delay_step"
     ]
 
     def __init__(self, kernel_weights, kernel_shape=None, strides=None,
                  padding=None, pool_shape=None, pool_stride=None,
                  positive_receptor_type="excitatory",
-                 negative_receptor_type="inhibitory", safe=True,
+                 negative_receptor_type="inhibitory",
+                 horizontal_delay_step=0,
+                 safe=True,
                  verbose=False, callback=None):
         """
         :param kernel_weights:
@@ -134,6 +138,8 @@ def __init__(self, kernel_weights, kernel_shape=None, strides=None,
         self.__positive_receptor_type = positive_receptor_type
         self.__negative_receptor_type = negative_receptor_type
 
+        self.__horizontal_delay_step = horizontal_delay_step
+
     @property
     def positive_receptor_type(self):
         return self.__positive_receptor_type
@@ -215,11 +221,10 @@ def get_post_shape(self, shape):
             shape = (post_pool_shape // self.__pool_stride) + 1
 
         kernel_shape = numpy.array(self.__kernel_weights.shape)
-        post_shape = (shape - (kernel_shape - 1) +
-                      (2 * self.__padding_shape))
+        post_shape = shape - kernel_shape + (2 * self.__padding_shape)
 
         return numpy.clip(
-            post_shape // self.__strides, 1, numpy.inf).astype('int')
+            post_shape // self.__strides + 1, 1, numpy.inf).astype('int')
 
     @overrides(AbstractConnector.validate_connection)
     def validate_connection(self, application_edge, synapse_info):
@@ -230,7 +235,9 @@ def validate_connection(self, application_edge, synapse_info):
                 "The ConvolutionConnector only works where the Populations"
                 " of a Projection are both 2D.  Please ensure that both the"
                 " Populations use a Grid2D structure.")
-        expected_post_shape = tuple(self.get_post_shape(pre.atoms_shape))
+        pre_shape = pre.atoms_shape
+        expected_post_shape = tuple(self.get_post_shape((pre_shape[1], pre_shape[0])))
+        expected_post_shape = expected_post_shape[1], expected_post_shape[0]
         if expected_post_shape != post.atoms_shape:
             raise ConfigurationException(
                 f"With a source population with shape {pre.atoms_shape}, "
@@ -281,10 +288,20 @@ def get_connected_vertices(self, s_info, source_vertex, target_vertex):
         pre_slices = [m_vertex.vertex_slice for m_vertex in pre_vertices]
         pre_slices_x = [vtx_slice.get_slice(0) for vtx_slice in pre_slices]
         pre_slices_y = [vtx_slice.get_slice(1) for vtx_slice in pre_slices]
-        pre_ranges = [[[px.start, py.start], [px.stop - 1, py.stop - 1]]
+        pre_ranges = [[[py.start, px.start], [py.stop - 1, px.stop - 1]]
                       for px, py in zip(pre_slices_x, pre_slices_y)]
-        pres_as_posts = self.__pre_as_post(pre_ranges)
-        hlf_k_w, hlf_k_h = numpy.array(self.__kernel_weights.shape) // 2
+        pre_vertex_in_post_layer, start_i = self.__pre_as_post(pre_ranges)
+
+        pre_vertex_in_post_layer_upper_left = pre_vertex_in_post_layer[:,0]
+        pre_vertex_in_post_layer_lower_right = pre_vertex_in_post_layer[:,1]
+
+        kernel_shape = numpy.array(self.__kernel_weights.shape)
+
+        j = (kernel_shape - 1 - start_i) // self.__strides
+        j_upper_left = j[:,0]
+
+        pre_vertex_max_reach_in_post_layer_upper_left = pre_vertex_in_post_layer_upper_left - j_upper_left
+        pre_vertex_max_reach_in_post_layer_lower_right = pre_vertex_in_post_layer_lower_right
 
         connected = list()
         for post in target_vertex.splitter.get_in_coming_vertices(
@@ -293,18 +310,18 @@ def get_connected_vertices(self, s_info, source_vertex, target_vertex):
             post_slice_x = post_slice.get_slice(0)
             post_slice_y = post_slice.get_slice(1)
 
-            # Get ranges allowed in post
-            min_x = post_slice_x.start - hlf_k_w
-            max_x = (post_slice_x.stop + hlf_k_w) - 1
-            min_y = post_slice_y.start - hlf_k_h
-            max_y = (post_slice_y.stop + hlf_k_h) - 1
+            # Get ranges allowed in post vertex
+            min_x = post_slice_x.start
+            max_x = post_slice_x.stop - 1
+            min_y = post_slice_y.start
+            max_y = post_slice_y.stop - 1
 
             # Test that the start coords are in range i.e. less than max
             start_in_range = numpy.logical_not(
-                numpy.any(pres_as_posts[:, 0] > [max_x, max_y], axis=1))
+                numpy.any(pre_vertex_max_reach_in_post_layer_upper_left > [max_y, max_x], axis=1))
             # Test that the end coords are in range i.e. more than min
             end_in_range = numpy.logical_not(
-                numpy.any(pres_as_posts[:, 1] < [min_x, min_y], axis=1))
+                numpy.any(pre_vertex_max_reach_in_post_layer_lower_right < [min_y, min_x], axis=1))
             # When both things are true, we have a vertex in range
             pre_in_range = pre_vertices[
                 numpy.logical_and(start_in_range, end_in_range)]
@@ -315,17 +332,18 @@ def get_connected_vertices(self, s_info, source_vertex, target_vertex):
     def __pre_as_post(self, pre_coords):
         """ Write pre coords as post coords.
 
-        :param Iterable pre_coords: An iterable of (x, y) coordinates
+        :param Iterable pre_coords: An iterable of (y, x) coordinates
         :rtype: numpy.ndarray
         """
         coords = numpy.array(pre_coords)
         if self.__pool_stride is not None:
             coords //= self.__pool_stride
 
-        kernel_shape = numpy.array(self.__kernel_weights.shape)
-        coords = coords - kernel_shape // 2 + self.__padding_shape
-        coords //= self.__strides
-        return coords
+        coords += self.__padding_shape
+        coord_by_strides = coords // self.__strides
+        start_i = coords % self.__strides
+
+        return coord_by_strides, start_i
 
     @property
     def local_only_n_bytes(self):
@@ -343,9 +361,9 @@ def write_local_only_data(
             weight_scales):
         # Get info about things
         kernel_shape = self.__kernel_weights.shape
-        ps_x, ps_y = 1, 1
+        ps_y, ps_x = 1, 1
         if self.__pool_stride is not None:
-            ps_x, ps_y = self.__pool_stride
+            ps_y, ps_x = self.__pool_stride
 
         # Write source key info
         spec.write_value(key, data_type=DataType.UINT32)
@@ -376,13 +394,17 @@ def write_local_only_data(
         # Write remaining connector details
         spec.write_value(start[1], data_type=DataType.INT16)
         spec.write_value(start[0], data_type=DataType.INT16)
-        spec.write_value(kernel_shape[1], data_type=DataType.INT16)
         spec.write_value(kernel_shape[0], data_type=DataType.INT16)
-        spec.write_value(self.__padding_shape[1], data_type=DataType.INT16)
+        spec.write_value(kernel_shape[1], data_type=DataType.INT16)
         spec.write_value(self.__padding_shape[0], data_type=DataType.INT16)
+        spec.write_value(self.__padding_shape[1], data_type=DataType.INT16)
+        spec.write_value(self.__recip(self.__strides[0]),
+                         data_type=DataType.INT16)
         spec.write_value(self.__recip(self.__strides[1]),
                          data_type=DataType.INT16)
-        spec.write_value(self.__recip(self.__strides[0]),
+        spec.write_value(self.__strides[0],
+                         data_type=DataType.INT16)
+        spec.write_value(self.__strides[1],
                          data_type=DataType.INT16)
         spec.write_value(self.__recip(ps_y), data_type=DataType.INT16)
         spec.write_value(self.__recip(ps_x), data_type=DataType.INT16)
@@ -395,6 +417,12 @@ def write_local_only_data(
         spec.write_value(pos_synapse_type, data_type=DataType.UINT16)
         spec.write_value(neg_synapse_type, data_type=DataType.UINT16)
 
+        # Write delay
+        spec.write_value(app_edge.post_vertex.synapse_dynamics.delay *
+                         SpynnakerDataView.get_simulation_time_step_per_ms())
+
+        spec.write_value(self.__horizontal_delay_step, data_type=DataType.UINT32)
+
         # Encode weights with weight scaling
         encoded_kernel_weights = self.__kernel_weights.flatten()
         if len(encoded_kernel_weights) % 2 != 0:
diff --git a/spynnaker/pyNN/models/neural_projections/connectors/pool_dense_connector.py b/spynnaker/pyNN/models/neural_projections/connectors/pool_dense_connector.py
index bc37b158290..04d3ce5c14a 100644
--- a/spynnaker/pyNN/models/neural_projections/connectors/pool_dense_connector.py
+++ b/spynnaker/pyNN/models/neural_projections/connectors/pool_dense_connector.py
@@ -25,11 +25,12 @@
 from collections.abc import Iterable
 from spinn_front_end_common.utilities.exceptions import ConfigurationException
 from spynnaker.pyNN.models.abstract_models import HasShapeKeyFields
+from spynnaker.pyNN.data.spynnaker_data_view import SpynnakerDataView
 
 
 _DIMENSION_SIZE = (2 * BYTES_PER_WORD) + (6 * BYTES_PER_SHORT)
 _KEY_INFO_SIZE = 3 * BYTES_PER_WORD
-_CONN_SIZE = _KEY_INFO_SIZE + (2 * BYTES_PER_WORD) + (2 * BYTES_PER_SHORT)
+_CONN_SIZE = _KEY_INFO_SIZE + (3 * BYTES_PER_WORD) + (2 * BYTES_PER_SHORT)
 _DIM_DTYPE = [("mask", "uint32"), ("shift", "uint32"), ("pre_start", "uint16"),
               ("pre_in_post_start", "uint16"), ("pre_in_post_end", "uint16"),
               ("pre_in_post_shape", "uint16"), ("recip_pool_stride", "uint16"),
@@ -283,6 +284,10 @@ def write_local_only_data(
         spec.write_value(pos_synapse_type, data_type=DataType.UINT16)
         spec.write_value(neg_synapse_type, data_type=DataType.UINT16)
 
+        # Write delay
+        spec.write_value(app_edge.post_vertex.synapse_dynamics.delay *
+                         SpynnakerDataView.get_simulation_time_step_per_ms())
+
         # Generate the dimension information
         dim_info = numpy.zeros(n_dims, dtype=_DIM_DTYPE)
         if self.__pool_stride is not None:
diff --git a/spynnaker/pyNN/models/neuron/local_only/local_only_convolution.py b/spynnaker/pyNN/models/neuron/local_only/local_only_convolution.py
index 7e27e8e2787..60c4eb11449 100644
--- a/spynnaker/pyNN/models/neuron/local_only/local_only_convolution.py
+++ b/spynnaker/pyNN/models/neuron/local_only/local_only_convolution.py
@@ -38,14 +38,20 @@ class LocalOnlyConvolution(AbstractLocalOnly, AbstractSupportsSignedWeights):
     """
 
     __slots__ = [
-        "__cached_2d_overlaps"
+        "__cached_2d_overlaps",
+        "__delay"
     ]
 
-    def __init__(self):
+    def __init__(self, delay=None):
+        """
+        :param float delay:
+            The delay used in the connection; by default 1 time step
+        """
         # Store the overlaps between 2d vertices to avoid recalculation
         self.__cached_2d_overlaps = dict()
-
-        # Store the merged keys for sources to avoid recalculation
+        self.__delay = delay
+        if delay is None:
+            self.__delay = SpynnakerDataView.get_simulation_time_step_ms()
 
     @overrides(AbstractLocalOnly.merge)
     def merge(self, synapse_dynamics):
@@ -193,7 +199,7 @@ def __get_rinfo_for_sources(self, key_cache, srcs, incoming):
     @property
     @overrides(AbstractLocalOnly.delay)
     def delay(self):
-        return SpynnakerDataView.get_simulation_time_step_ms()
+        return self.__delay
 
     @property
     @overrides(AbstractLocalOnly.weight)
diff --git a/spynnaker/pyNN/models/neuron/local_only/local_only_pool_dense.py b/spynnaker/pyNN/models/neuron/local_only/local_only_pool_dense.py
index da63852a5d1..4f8c7b2e58f 100644
--- a/spynnaker/pyNN/models/neuron/local_only/local_only_pool_dense.py
+++ b/spynnaker/pyNN/models/neuron/local_only/local_only_pool_dense.py
@@ -31,6 +31,17 @@ class LocalOnlyPoolDense(AbstractLocalOnly, AbstractSupportsSignedWeights):
     """ A convolution synapse dynamics that can process spikes with only DTCM
     """
 
+    __slots__ = ["__delay"]
+
+    def __init__(self, delay=None):
+        """
+        :param float delay:
+            The delay used in the connection; by default 1 time step
+        """
+        self.__delay = delay
+        if delay is None:
+            self.__delay = SpynnakerDataView.get_simulation_time_step_ms()
+
     @overrides(AbstractLocalOnly.merge)
     def merge(self, synapse_dynamics):
         if not isinstance(synapse_dynamics, LocalOnlyPoolDense):
@@ -135,7 +146,7 @@ def __merge_key_and_mask(self, key_a, mask_a, key_b, mask_b):
     @property
     @overrides(AbstractLocalOnly.delay)
     def delay(self):
-        return SpynnakerDataView.get_simulation_time_step_ms()
+        return self.__delay
 
     @property
     @overrides(AbstractLocalOnly.weight)
diff --git a/spynnaker/pyNN/models/neuron/population_machine_local_only_combined_vertex.py b/spynnaker/pyNN/models/neuron/population_machine_local_only_combined_vertex.py
index d4f9cbe0ff5..1ce4ac59463 100644
--- a/spynnaker/pyNN/models/neuron/population_machine_local_only_combined_vertex.py
+++ b/spynnaker/pyNN/models/neuron/population_machine_local_only_combined_vertex.py
@@ -25,6 +25,7 @@
 from .population_machine_common import CommonRegions, PopulationMachineCommon
 from .population_machine_neurons import (
     NeuronRegions, PopulationMachineNeurons, NeuronProvenance)
+from spynnaker.pyNN.data.spynnaker_data_view import SpynnakerDataView
 
 
 class LocalOnlyProvenance(ctypes.LittleEndianStructure):
@@ -270,12 +271,15 @@ def __write_local_only_data(self, spec):
         log_n_max_atoms = get_n_bits(self._max_atoms_per_core)
         log_n_synapse_types = get_n_bits(
             self._app_vertex.neuron_impl.get_n_synapse_types())
-        # Delay is always 1
-        log_max_delay = 1
+        # Find the maximum delay
+        # pylint: disable=protected-access
+        max_delay = SpynnakerDataView.get_simulation_time_step_per_ms() * max(
+            proj._synapse_information.delays
+            for proj in self._app_vertex.incoming_projections)
 
         spec.write_value(log_n_max_atoms)
         spec.write_value(log_n_synapse_types)
-        spec.write_value(log_max_delay)
+        spec.write_value(get_n_bits(max_delay))
         spec.write_value(self._app_vertex.incoming_spike_buffer_size)
         spec.write_value(int(self._app_vertex.drop_late_spikes))