diff --git a/src/db/db/dbDeepRegion.cc b/src/db/db/dbDeepRegion.cc
index a4100ef19..c869ac5af 100644
--- a/src/db/db/dbDeepRegion.cc
+++ b/src/db/db/dbDeepRegion.cc
@@ -1070,56 +1070,163 @@ DeepRegion::snapped (db::Coord gx, db::Coord gy)
   return res.release ();
 }
 
+namespace
+{
+
+class PolygonToEdgeLocalOperation
+  : public local_operation<db::PolygonRef, db::PolygonRef, db::Edge>
+{
+public:
+  PolygonToEdgeLocalOperation ()
+    : local_operation<db::PolygonRef, db::PolygonRef, db::Edge> ()
+  {
+    //  .. nothing yet ..
+  }
+
+  virtual db::Coord dist () const { return 1; }
+  virtual bool requests_single_subjects () const { return true; }
+  virtual std::string description () const { return std::string ("polygon to edges"); }
+
+  virtual void do_compute_local (db::Layout * /*layout*/, const shape_interactions<db::PolygonRef, db::PolygonRef> &interactions, std::vector<std::unordered_set<db::Edge> > &results, size_t /*max_vertex_count*/, double /*area_ratio*/) const
+  {
+    db::EdgeProcessor ep;
+
+    for (shape_interactions<db::PolygonRef, db::PolygonRef>::subject_iterator s = interactions.begin_subjects (); s != interactions.end_subjects (); ++s) {
+      ep.insert (s->second);
+    }
+
+    if (interactions.num_intruders () == 0) {
+
+      db::EdgeToEdgeSetGenerator eg (results.front ());
+      db::MergeOp op (0);
+      ep.process (eg, op);
+
+    } else {
+
+      //  With intruders: to compute our local contribution we take the edges without and with intruders
+      //  and deliver what is in both sets
+
+      db::MergeOp op (0);
+
+      std::vector<Edge> edges1;
+      db::EdgeContainer ec1 (edges1);
+      ep.process (ec1, op);
+
+      ep.clear ();
+
+      for (shape_interactions<db::PolygonRef, db::PolygonRef>::subject_iterator s = interactions.begin_subjects (); s != interactions.end_subjects (); ++s) {
+        ep.insert (s->second);
+      }
+      for (shape_interactions<db::PolygonRef, db::PolygonRef>::intruder_iterator i = interactions.begin_intruders (); i != interactions.end_intruders (); ++i) {
+        ep.insert (i->second.second);
+      }
+
+      std::vector<Edge> edges2;
+      db::EdgeContainer ec2 (edges2);
+      ep.process (ec2, op);
+
+      //  Runs the boolean AND between the result with and without intruders
+
+      db::box_scanner<db::Edge, size_t> scanner;
+      scanner.reserve (edges1.size () + edges2.size ());
+
+      for (std::vector<Edge>::const_iterator i = edges1.begin (); i != edges1.end (); ++i) {
+        scanner.insert (i.operator-> (), 0);
+      }
+      for (std::vector<Edge>::const_iterator i = edges2.begin (); i != edges2.end (); ++i) {
+        scanner.insert (i.operator-> (), 1);
+      }
+
+      EdgeBooleanClusterCollector<std::unordered_set<db::Edge> > cluster_collector (&results.front (), EdgeAnd);
+      scanner.process (cluster_collector, 1, db::box_convert<db::Edge> ());
+
+    }
+
+  }
+};
+
+}
+
 EdgesDelegate *
 DeepRegion::edges (const EdgeFilterBase *filter) const
 {
-  const db::DeepLayer &polygons = merged_deep_layer ();
+  if (! filter && merged_semantics ()) {
 
-  std::unique_ptr<VariantsCollectorBase> vars;
+    //  Hierarchical edge detector - no pre-merge required
 
-  if (filter && filter->vars ()) {
+    const db::DeepLayer &polygons = deep_layer ();
 
-    vars.reset (new db::VariantsCollectorBase (filter->vars ()));
+    db::PolygonToEdgeLocalOperation op;
 
-    vars->collect (polygons.layout (), polygons.initial_cell ());
+    db::local_processor<db::PolygonRef, db::PolygonRef, db::Edge> proc (const_cast<db::Layout *> (&polygons.layout ()),
+                                                                        const_cast<db::Cell *> (&polygons.initial_cell ()),
+                                                                        polygons.breakout_cells ());
 
-    //  NOTE: m_merged_polygons is mutable, so why is the const_cast needed?
-    const_cast<db::DeepLayer &> (polygons).separate_variants (*vars);
+    proc.set_description (progress_desc ());
+    proc.set_report_progress (report_progress ());
+    proc.set_base_verbosity (base_verbosity ());
+    proc.set_threads (polygons.store ()->threads ());
 
-  }
+    //  a boolean core makes somewhat better hierarchy
+    proc.set_boolean_core (true);
 
-  db::Layout &layout = const_cast<db::Layout &> (polygons.layout ());
+    std::unique_ptr<db::DeepEdges> res (new db::DeepEdges (polygons.derived ()));
 
-  std::unique_ptr<db::DeepEdges> res (new db::DeepEdges (polygons.derived ()));
-  for (db::Layout::iterator c = layout.begin (); c != layout.end (); ++c) {
+    proc.run (&op, polygons.layer (), foreign_idlayer (), res->deep_layer ().layer ());
+
+    return res.release ();
+
+  } else {
+
+    const db::DeepLayer &polygons = merged_deep_layer ();
+
+    std::unique_ptr<VariantsCollectorBase> vars;
+
+    if (filter && filter->vars ()) {
+
+      vars.reset (new db::VariantsCollectorBase (filter->vars ()));
+
+      vars->collect (polygons.layout (), polygons.initial_cell ());
+
+      //  NOTE: m_merged_polygons is mutable, so why is the const_cast needed?
+      const_cast<db::DeepLayer &> (polygons).separate_variants (*vars);
 
-    db::ICplxTrans tr;
-    if (vars.get ()) {
-      const std::map<db::ICplxTrans, size_t> &v = vars->variants (c->cell_index ());
-      tl_assert (v.size () == size_t (1));
-      tr = v.begin ()->first;
     }
 
-    const db::Shapes &s = c->shapes (polygons.layer ());
-    db::Shapes &st = c->shapes (res->deep_layer ().layer ());
+    db::Layout &layout = const_cast<db::Layout &> (polygons.layout ());
 
-    for (db::Shapes::shape_iterator si = s.begin (db::ShapeIterator::All); ! si.at_end (); ++si) {
+    std::unique_ptr<db::DeepEdges> res (new db::DeepEdges (polygons.derived ()));
+    for (db::Layout::iterator c = layout.begin (); c != layout.end (); ++c) {
 
-      db::Polygon poly;
-      si->polygon (poly);
+      db::ICplxTrans tr;
+      if (vars.get ()) {
+        const std::map<db::ICplxTrans, size_t> &v = vars->variants (c->cell_index ());
+        tl_assert (v.size () == size_t (1));
+        tr = v.begin ()->first;
+      }
 
-      for (db::Polygon::polygon_edge_iterator e = poly.begin_edge (); ! e.at_end (); ++e) {
-        if (! filter || filter->selected ((*e).transformed (tr))) {
-          st.insert (*e);
+      const db::Shapes &s = c->shapes (polygons.layer ());
+      db::Shapes &st = c->shapes (res->deep_layer ().layer ());
+
+      for (db::Shapes::shape_iterator si = s.begin (db::ShapeIterator::All); ! si.at_end (); ++si) {
+
+        db::Polygon poly;
+        si->polygon (poly);
+
+        for (db::Polygon::polygon_edge_iterator e = poly.begin_edge (); ! e.at_end (); ++e) {
+          if (! filter || filter->selected ((*e).transformed (tr))) {
+            st.insert (*e);
+          }
         }
+
       }
 
     }
 
-  }
+    res->set_is_merged (merged_semantics () || is_merged ());
+    return res.release ();
 
-  res->set_is_merged (merged_semantics () || is_merged ());
-  return res.release ();
+  }
 }
 
 RegionDelegate *
diff --git a/src/db/db/dbHierProcessor.cc b/src/db/db/dbHierProcessor.cc
index df90ff1ba..8d39b1a00 100644
--- a/src/db/db/dbHierProcessor.cc
+++ b/src/db/db/dbHierProcessor.cc
@@ -409,6 +409,16 @@ local_processor_cell_contexts<TS, TI, TR>::create (const context_key_type &intru
   return &m_contexts[intruders];
 }
 
+template <class TR>
+static void
+subtract_set (std::unordered_set<TR> &res, const std::unordered_set<TR> &other)
+{
+  //  for everything else, we don't use a boolean core but just set intersection
+  for (typename std::unordered_set<TR>::const_iterator o = other.begin (); o != other.end (); ++o) {
+    res.erase (*o);
+  }
+}
+
 template <class TS, class TI>
 static void
 subtract (std::unordered_set<db::PolygonRef> &res, const std::unordered_set<db::PolygonRef> &other, db::Layout *layout, const db::local_processor<TS, TI, db::PolygonRef> *proc)
@@ -417,6 +427,11 @@ subtract (std::unordered_set<db::PolygonRef> &res, const std::unordered_set<db::
     return;
   }
 
+  if (! proc->boolean_core ()) {
+    subtract_set (res, other);
+    return;
+  }
+
   size_t max_vertex_count = proc->max_vertex_count ();
   double area_ratio = proc->area_ratio ();
 
@@ -449,14 +464,60 @@ subtract (std::unordered_set<db::PolygonRef> &res, const std::unordered_set<db::
   ep.process (pg, op);
 }
 
+template <class TS, class TI>
+static void
+subtract (std::unordered_set<db::Edge> &res, const std::unordered_set<db::Edge> &other, db::Layout *layout, const db::local_processor<TS, TI, db::Edge> *proc)
+{
+  if (other.empty ()) {
+    return;
+  }
+
+  if (! proc->boolean_core ()) {
+    subtract_set (res, other);
+    return;
+  }
+
+  db::box_scanner<db::Edge, size_t> scanner;
+  scanner.reserve (res.size () + other.size ());
+
+  for (std::unordered_set<Edge>::const_iterator i = res.begin (); i != res.end (); ++i) {
+    scanner.insert (i.operator-> (), 0);
+  }
+  for (std::unordered_set<Edge>::const_iterator i = other.begin (); i != other.end (); ++i) {
+    scanner.insert (i.operator-> (), 1);
+  }
+
+  std::unordered_set<db::Edge> result;
+  EdgeBooleanClusterCollector<std::unordered_set<db::Edge> > cluster_collector (&result, EdgeNot);
+  scanner.process (cluster_collector, 1, db::box_convert<db::Edge> ());
+
+  res.swap (result);
+}
+
 template <class TS, class TI, class TR>
 static void
 subtract (std::unordered_set<TR> &res, const std::unordered_set<TR> &other, db::Layout * /*layout*/, const db::local_processor<TS, TI, TR> * /*proc*/)
 {
-  //  for edges, we don't use a boolean core but just set intersection
-  for (typename std::unordered_set<TR>::const_iterator o = other.begin (); o != other.end (); ++o) {
-    res.erase (*o);
-  }
+  subtract_set (res, other);
+}
+
+//  determines the default boolean core flag per result type
+
+namespace
+{
+
+template <class TR>
+struct default_boolean_core
+{
+  bool operator() () const { return false; }
+};
+
+template <>
+struct default_boolean_core<db::PolygonRef>
+{
+  bool operator() () const { return true; }
+};
+
 }
 
 namespace {
@@ -1221,7 +1282,7 @@ local_processor<TS, TI, TR>::local_processor (db::Layout *layout, db::Cell *top,
   : mp_subject_layout (layout), mp_intruder_layout (layout),
     mp_subject_top (top), mp_intruder_top (top),
     mp_subject_breakout_cells (breakout_cells), mp_intruder_breakout_cells (breakout_cells),
-    m_report_progress (true), m_nthreads (0), m_max_vertex_count (0), m_area_ratio (0.0), m_base_verbosity (30), m_progress (0), mp_progress (0)
+    m_report_progress (true), m_nthreads (0), m_max_vertex_count (0), m_area_ratio (0.0), m_boolean_core (default_boolean_core<TR> () ()), m_base_verbosity (30), m_progress (0), mp_progress (0)
 {
   //  .. nothing yet ..
 }
@@ -1231,7 +1292,7 @@ local_processor<TS, TI, TR>::local_processor (db::Layout *subject_layout, db::Ce
   : mp_subject_layout (subject_layout), mp_intruder_layout (intruder_layout),
     mp_subject_top (subject_top), mp_intruder_top (intruder_top),
     mp_subject_breakout_cells (subject_breakout_cells), mp_intruder_breakout_cells (intruder_breakout_cells),
-    m_report_progress (true), m_nthreads (0), m_max_vertex_count (0), m_area_ratio (0.0), m_base_verbosity (30), m_progress (0), mp_progress (0)
+    m_report_progress (true), m_nthreads (0), m_max_vertex_count (0), m_area_ratio (0.0), m_boolean_core (default_boolean_core<TR> () ()), m_base_verbosity (30), m_progress (0), mp_progress (0)
 {
   //  .. nothing yet ..
 }
@@ -1535,6 +1596,11 @@ void local_processor<TS, TI, TR>::compute_contexts (local_processor_contexts<TS,
       scanner.process (rec, dist, inst_bcs, db::box_convert<TI> ());
     }
 
+    //  this cache should reduce the effort of checking array vs. array
+    typedef std::pair<unsigned int, std::pair<db::cell_index_type, db::ICplxTrans> > effective_instance_cache_key_type;
+    typedef std::map<effective_instance_cache_key_type, std::pair<bool, db::CellInstArray> > effective_instance_cache_type;
+    effective_instance_cache_type effective_instance_cache;
+
     for (typename std::unordered_map<const db::CellInstArray *, interaction_value_type>::const_iterator i = interactions.begin (); i != interactions.end (); ++i) {
 
       db::Cell &subject_child_cell = mp_subject_layout->cell (i->first->object ().cell_index ());
@@ -1568,17 +1634,28 @@ void local_processor<TS, TI, TR>::compute_contexts (local_processor_contexts<TS,
             db::box_convert <db::CellInst, true> inst_bcii (*mp_intruder_layout, ail);
 
             for (std::unordered_set<const db::CellInstArray *>::const_iterator j = i->second.first.begin (); j != i->second.first.end (); ++j) {
+
               for (db::CellInstArray::iterator k = (*j)->begin_touching (safe_box_enlarged (nbox, -1, -1), inst_bcii); ! k.at_end (); ++k) {
+
                 db::ICplxTrans tk = (*j)->complex_trans (*k);
                 //  NOTE: no self-interactions
                 if (i->first != *j || tn != tk) {
+
                   //  optimize the intruder instance so it will be as low as possible
-                  std::pair<bool, db::CellInstArray> ei = effective_instance (contexts.subject_layer (), i->first->object ().cell_index (), ail, (*j)->object ().cell_index (), tni * tk, dist);
-                  if (ei.first) {
-                    intruders_below.first.insert (ei.second);
+                  effective_instance_cache_key_type key (ail, std::make_pair ((*j)->object ().cell_index (), tni * tk));
+                  effective_instance_cache_type::iterator cached = effective_instance_cache.find (key);
+                  if (cached == effective_instance_cache.end ()) {
+                    std::pair<bool, db::CellInstArray> ei = effective_instance (contexts.subject_layer (), i->first->object ().cell_index (), ail, (*j)->object ().cell_index (), tni * tk, dist);
+                    cached = effective_instance_cache.insert (std::make_pair (key, ei)).first;
                   }
+                  if (cached->second.first) {
+                    intruders_below.first.insert (cached->second.second);
+                  }
+
                 }
+
               }
+
             }
 
           }
diff --git a/src/db/db/dbHierProcessor.h b/src/db/db/dbHierProcessor.h
index c65da727f..712ebf53e 100644
--- a/src/db/db/dbHierProcessor.h
+++ b/src/db/db/dbHierProcessor.h
@@ -479,6 +479,16 @@ public:
     return m_area_ratio;
   }
 
+  void set_boolean_core (double boolean_core)
+  {
+    m_boolean_core = boolean_core;
+  }
+
+  double boolean_core () const
+  {
+    return m_boolean_core;
+  }
+
 private:
   template<typename, typename, typename> friend class local_processor_cell_contexts;
   template<typename, typename, typename> friend class local_processor_context_computation_task;
@@ -494,6 +504,7 @@ private:
   unsigned int m_nthreads;
   size_t m_max_vertex_count;
   double m_area_ratio;
+  bool m_boolean_core;
   int m_base_verbosity;
   mutable std::unique_ptr<tl::Job<local_processor_context_computation_worker<TS, TI, TR> > > mp_cc_job;
   mutable size_t m_progress;
diff --git a/src/db/unit_tests/dbDeepRegionTests.cc b/src/db/unit_tests/dbDeepRegionTests.cc
index f02a2f1db..bccfefbaa 100644
--- a/src/db/unit_tests/dbDeepRegionTests.cc
+++ b/src/db/unit_tests/dbDeepRegionTests.cc
@@ -800,7 +800,7 @@ TEST(13_Edges)
 
   db::Region r3 (db::RecursiveShapeIterator (ly, top_cell, l3), dss);
   db::Edges r3edges = r3.edges ();
-  EXPECT_EQ (r3edges.is_merged (), true);
+  EXPECT_EQ (r3edges.is_merged (), false);
 
   db::EdgeLengthFilter f (0, 500, true);
   db::Edges r3edges_filtered = r3.edges (f);
@@ -816,6 +816,46 @@ TEST(13_Edges)
   db::compare_layouts (_this, target, tl::testsrc () + "/testdata/algo/deep_region_au13.gds");
 }
 
+TEST(13b_Edges)
+{
+  db::Layout ly;
+  {
+    std::string fn (tl::testsrc ());
+    fn += "/testdata/algo/deep_region_edges.gds";
+    tl::InputStream stream (fn);
+    db::Reader reader (stream);
+    reader.read (ly);
+  }
+
+  db::cell_index_type top_cell_index = *ly.begin_top_down ();
+  db::Cell &top_cell = ly.cell (top_cell_index);
+
+  db::DeepShapeStore dss;
+
+  unsigned int l1 = ly.get_layer (db::LayerProperties (1, 0));
+  unsigned int l2 = ly.get_layer (db::LayerProperties (2, 0));
+
+  db::Region r1 (db::RecursiveShapeIterator (ly, top_cell, l1), dss);
+  db::Edges r1edges = r1.edges ();
+  EXPECT_EQ (r1edges.is_merged (), false);
+
+  db::Region r2 (db::RecursiveShapeIterator (ly, top_cell, l2), dss);
+  db::Edges r2edges = r2.edges ();
+  EXPECT_EQ (r2edges.is_merged (), false);
+
+
+  db::Layout target;
+  unsigned int target_top_cell_index = target.add_cell (ly.cell_name (top_cell_index));
+
+  target.insert (target_top_cell_index, target.get_layer (db::LayerProperties (1, 0)), r1);
+  target.insert (target_top_cell_index, target.get_layer (db::LayerProperties (2, 0)), r2);
+  target.insert (target_top_cell_index, target.get_layer (db::LayerProperties (11, 0)), r1edges);
+  target.insert (target_top_cell_index, target.get_layer (db::LayerProperties (12, 0)), r2edges);
+
+  CHECKPOINT();
+  db::compare_layouts (_this, target, tl::testsrc () + "/testdata/algo/deep_region_au13b.gds");
+}
+
 TEST(14_Interacting)
 {
   db::Layout ly;
@@ -936,7 +976,7 @@ TEST(14_Interacting)
     target.insert (target_top_cell_index, target.get_layer (db::LayerProperties (32, 0)), r6r.selected_interacting (r1er));
     target.insert (target_top_cell_index, target.get_layer (db::LayerProperties (33, 0)), r6r.selected_not_interacting (r1er));
 
-    EXPECT_EQ (r6.selected_interacting (r1e).is_merged (), true);
+    EXPECT_EQ (r6.selected_interacting (r1e).is_merged (), false);
     EXPECT_EQ (r6.selected_interacting (r1er).is_merged (), false);
     EXPECT_EQ (r6r.selected_interacting (r1e).is_merged (), false);
     EXPECT_EQ (r6r.selected_interacting (r1er).is_merged (), false);
diff --git a/testdata/algo/deep_edges_au10.gds b/testdata/algo/deep_edges_au10.gds
index 13ded556b..c809ae30c 100644
Binary files a/testdata/algo/deep_edges_au10.gds and b/testdata/algo/deep_edges_au10.gds differ
diff --git a/testdata/algo/deep_edges_au3.gds b/testdata/algo/deep_edges_au3.gds
index 62914f828..4bb15fdec 100644
Binary files a/testdata/algo/deep_edges_au3.gds and b/testdata/algo/deep_edges_au3.gds differ
diff --git a/testdata/algo/deep_edges_au4.gds b/testdata/algo/deep_edges_au4.gds
index f91ff8526..c9a752daa 100644
Binary files a/testdata/algo/deep_edges_au4.gds and b/testdata/algo/deep_edges_au4.gds differ
diff --git a/testdata/algo/deep_region_au13.gds b/testdata/algo/deep_region_au13.gds
index ade409f90..14c9e6e4e 100644
Binary files a/testdata/algo/deep_region_au13.gds and b/testdata/algo/deep_region_au13.gds differ
diff --git a/testdata/algo/deep_region_au13b.gds b/testdata/algo/deep_region_au13b.gds
new file mode 100644
index 000000000..98ccf1b00
Binary files /dev/null and b/testdata/algo/deep_region_au13b.gds differ
diff --git a/testdata/algo/deep_region_au14b.gds b/testdata/algo/deep_region_au14b.gds
index d02050199..fd5afab25 100644
Binary files a/testdata/algo/deep_region_au14b.gds and b/testdata/algo/deep_region_au14b.gds differ
diff --git a/testdata/algo/deep_region_au25b.gds b/testdata/algo/deep_region_au25b.gds
index 8d0ac52ad..5d0c39497 100644
Binary files a/testdata/algo/deep_region_au25b.gds and b/testdata/algo/deep_region_au25b.gds differ
diff --git a/testdata/algo/deep_region_edges.gds b/testdata/algo/deep_region_edges.gds
new file mode 100644
index 000000000..c0a9af034
Binary files /dev/null and b/testdata/algo/deep_region_edges.gds differ
diff --git a/testdata/drc/drcSimpleTests_au16.gds b/testdata/drc/drcSimpleTests_au16.gds
index c8350ab1b..7970f2426 100644
Binary files a/testdata/drc/drcSimpleTests_au16.gds and b/testdata/drc/drcSimpleTests_au16.gds differ