diff --git a/.github/scripts/check-commit.sh b/.github/scripts/check-commit.sh
index fe1de2bf..d45ac6ff 100755
--- a/.github/scripts/check-commit.sh
+++ b/.github/scripts/check-commit.sh
@@ -45,7 +45,7 @@ search () {
     done
 }
 
-submodules=("cva6" "boom" "ibex" "gemmini" "hwacha" "icenet" "nvdla" "rocket-chip" "sha3" "sifive-blocks" "sifive-cache" "testchipip" "riscv-sodor" "mempress")
+submodules=("cva6" "boom" "ibex" "gemmini" "hwacha" "icenet" "nvdla" "rocket-chip" "sha3" "sifive-blocks" "sifive-cache" "testchipip" "riscv-sodor" "mempress" "bar-fetchers" "shuttle")
 dir="generators"
 branches=("master" "main" "dev")
 search
diff --git a/.github/scripts/defaults.sh b/.github/scripts/defaults.sh
index 1c1b7def..d276a891 100755
--- a/.github/scripts/defaults.sh
+++ b/.github/scripts/defaults.sh
@@ -28,7 +28,7 @@ REMOTE_COURSIER_CACHE=$REMOTE_WORK_DIR/.coursier-cache
 
 # key value store to get the build groups
 declare -A grouping
-grouping["group-cores"]="chipyard-cva6 chipyard-ibex chipyard-rocket chipyard-hetero chipyard-boom chipyard-sodor chipyard-digitaltop chipyard-multiclock-rocket chipyard-nomem-scratchpad chipyard-spike chipyard-clone"
+grouping["group-cores"]="chipyard-cva6 chipyard-ibex chipyard-rocket chipyard-hetero chipyard-boom chipyard-sodor chipyard-digitaltop chipyard-multiclock-rocket chipyard-nomem-scratchpad chipyard-spike chipyard-clone chipyard-prefetchers chipyard-shuttle"
 grouping["group-peripherals"]="chipyard-dmirocket chipyard-dmiboom chipyard-spiflashwrite chipyard-mmios chipyard-nocores chipyard-manyperipherals chipyard-chiplike"
 grouping["group-accels"]="chipyard-mempress chipyard-sha3 chipyard-hwacha chipyard-gemmini chipyard-manymmioaccels chipyard-nvdla"
 grouping["group-constellation"]="chipyard-constellation"
@@ -42,9 +42,10 @@ mapping["chipyard-rocket"]=" CONFIG=QuadChannelRocketConfig"
 mapping["chipyard-dmirocket"]=" CONFIG=dmiRocketConfig"
 mapping["chipyard-sha3"]=" CONFIG=Sha3RocketConfig"
 mapping["chipyard-mempress"]=" CONFIG=MempressRocketConfig"
+mapping["chipyard-prefetchers"]=" CONFIG=PrefetchingRocketConfig"
 mapping["chipyard-digitaltop"]=" TOP=DigitalTop"
 mapping["chipyard-manymmioaccels"]=" CONFIG=ManyMMIOAcceleratorRocketConfig"
-mapping["chipyard-nvdla"]=" CONFIG=SmallNVDLARocketConfig"
+mapping["chipyard-nvdla"]=" CONFIG=SmallNVDLARocketConfig verilog"
 mapping["chipyard-hetero"]=" CONFIG=LargeBoomAndRocketConfig"
 mapping["chipyard-boom"]=" CONFIG=MediumBoomCosimConfig"
 mapping["chipyard-dmiboom"]=" CONFIG=dmiMediumBoomCosimConfig"
@@ -61,6 +62,7 @@ mapping["chipyard-nocores"]=" CONFIG=NoCoresConfig verilog"
 mapping["tracegen"]=" CONFIG=NonBlockingTraceGenL2Config"
 mapping["tracegen-boom"]=" CONFIG=BoomTraceGenConfig"
 mapping["chipyard-sodor"]=" CONFIG=Sodor5StageConfig"
+mapping["chipyard-shuttle"]=" CONFIG=ShuttleConfig"
 mapping["chipyard-multiclock-rocket"]=" CONFIG=MulticlockRocketConfig"
 mapping["chipyard-nomem-scratchpad"]=" CONFIG=MMIOScratchpadOnlyRocketConfig"
 mapping["chipyard-constellation"]=" CONFIG=SharedNoCConfig"
diff --git a/.github/scripts/run-tests.sh b/.github/scripts/run-tests.sh
index 4a5a0802..ac81411b 100755
--- a/.github/scripts/run-tests.sh
+++ b/.github/scripts/run-tests.sh
@@ -46,6 +46,9 @@ case $1 in
     chipyard-boom)
         run_bmark ${mapping[$1]}
         ;;
+    chipyard-shuttle)
+        run_bmark ${mapping[$1]}
+        ;;
     chipyard-dmiboom)
         $LOCAL_CHIPYARD_DIR/scripts/generate-ckpt.sh -b $RISCV/riscv64-unknown-elf/share/riscv-tests/benchmarks/dhrystone.riscv -i 10000
         make -C $LOCAL_SIM_DIR $DISABLE_SIM_PREREQ ${mapping[$1]} run-binary LOADARCH=$PWD/dhrystone.riscv.0x80000000.10000.loadarch
@@ -56,6 +59,9 @@ case $1 in
     chipyard-hetero)
         run_bmark ${mapping[$1]}
         ;;
+    chipyard-prefetchers)
+        make -C $LOCAL_SIM_DIR $DISABLE_SIM_PREREQ ${mapping[$1]} run-binary BINARY=$RISCV/riscv64-unknown-elf/share/riscv-tests/benchmarks/dhrystone.riscv
+        ;;
     rocketchip)
         run_bmark ${mapping[$1]}
         ;;
diff --git a/.github/workflows/chipyard-run-tests.yml b/.github/workflows/chipyard-run-tests.yml
index 02de6935..306d38b4 100644
--- a/.github/workflows/chipyard-run-tests.yml
+++ b/.github/workflows/chipyard-run-tests.yml
@@ -442,6 +442,29 @@ jobs:
           group-key: "group-cores"
           project-key: "chipyard-rocket"
 
+  chipyard-prefetchers-run-tests:
+    name: chipyard-prefetchers-run-tests
+    needs: prepare-chipyard-cores
+    runs-on: self-hosted
+    steps:
+      - name: Delete old checkout
+        run: |
+            ls -alh .
+            rm -rf ${{ github.workspace }}/* || true
+            rm -rf ${{ github.workspace }}/.* || true
+            ls -alh .
+      - name: Checkout
+        uses: actions/checkout@v3
+      - name: Git workaround
+        uses: ./.github/actions/git-workaround
+      - name: Create conda env
+        uses: ./.github/actions/create-conda-env
+      - name: Run tests
+        uses: ./.github/actions/run-tests
+        with:
+          group-key: "group-cores"
+          project-key: "chipyard-prefetchers"
+
   chipyard-hetero-run-tests:
     name: chipyard-hetero-run-tests
     needs: prepare-chipyard-cores
@@ -488,6 +511,29 @@ jobs:
           group-key: "group-cores"
           project-key: "chipyard-boom"
 
+  chipyard-shuttle-run-tests:
+    name: chipyard-shuttle-run-tests
+    needs: prepare-chipyard-cores
+    runs-on: self-hosted
+    steps:
+      - name: Delete old checkout
+        run: |
+            ls -alh .
+            rm -rf ${{ github.workspace }}/* || true
+            rm -rf ${{ github.workspace }}/.* || true
+            ls -alh .
+      - name: Checkout
+        uses: actions/checkout@v3
+      - name: Git workaround
+        uses: ./.github/actions/git-workaround
+      - name: Create conda env
+        uses: ./.github/actions/create-conda-env
+      - name: Run tests
+        uses: ./.github/actions/run-tests
+        with:
+          group-key: "group-cores"
+          project-key: "chipyard-shuttle"
+
   chipyard-cva6-run-tests:
     name: chipyard-cva6-run-tests
     needs: prepare-chipyard-cores
@@ -741,28 +787,28 @@ jobs:
           group-key: "group-accels"
           project-key: "chipyard-manymmioaccels"
 
-  chipyard-nvdla-run-tests:
-    name: chipyard-nvdla-run-tests
-    needs: prepare-chipyard-accels
-    runs-on: self-hosted
-    steps:
-      - name: Delete old checkout
-        run: |
-            ls -alh .
-            rm -rf ${{ github.workspace }}/* || true
-            rm -rf ${{ github.workspace }}/.* || true
-            ls -alh .
-      - name: Checkout
-        uses: actions/checkout@v3
-      - name: Git workaround
-        uses: ./.github/actions/git-workaround
-      - name: Create conda env
-        uses: ./.github/actions/create-conda-env
-      - name: Run tests
-        uses: ./.github/actions/run-tests
-        with:
-          group-key: "group-accels"
-          project-key: "chipyard-nvdla"
+  # chipyard-nvdla-run-tests:
+  #   name: chipyard-nvdla-run-tests
+  #   needs: prepare-chipyard-accels
+  #   runs-on: self-hosted
+  #   steps:
+  #     - name: Delete old checkout
+  #       run: |
+  #           ls -alh .
+  #           rm -rf ${{ github.workspace }}/* || true
+  #           rm -rf ${{ github.workspace }}/.* || true
+  #           ls -alh .
+  #     - name: Checkout
+  #       uses: actions/checkout@v3
+  #     - name: Git workaround
+  #       uses: ./.github/actions/git-workaround
+  #     - name: Create conda env
+  #       uses: ./.github/actions/create-conda-env
+  #     - name: Run tests
+  #       uses: ./.github/actions/run-tests
+  #       with:
+  #         group-key: "group-accels"
+  #         project-key: "chipyard-nvdla"
 
   chipyard-mempress-run-tests:
     name: chipyard-mempress-run-tests
@@ -1026,6 +1072,7 @@ jobs:
             chipyard-rocket-run-tests,
             chipyard-hetero-run-tests,
             chipyard-boom-run-tests,
+            chipyard-shuttle-run-tests,
             chipyard-cva6-run-tests,
             chipyard-ibex-run-tests,
             chipyard-sodor-run-tests,
@@ -1035,8 +1082,8 @@ jobs:
             chipyard-manyperipherals-run-tests,
             chipyard-sha3-run-tests,
             chipyard-gemmini-run-tests,
-            chipyard-manymmioaccels-run-tests,
-            chipyard-nvdla-run-tests,
+            chipyard-manymmioaccels-run-tests, # chipyard-nvdla-run-tests,
+            chipyard-prefetchers-run-tests,
             chipyard-mempress-run-tests,
             chipyard-constellation-run-tests,
             tracegen-boom-run-tests,
diff --git a/.gitignore b/.gitignore
index ac8e84d0..5dd75bec 100644
--- a/.gitignore
+++ b/.gitignore
@@ -27,3 +27,4 @@ project/project/
 .ivy2
 .sbt
 .classpath_cache/
+.vscode/
diff --git a/.gitmodules b/.gitmodules
index 5a3eb6fc..43f705fc 100644
--- a/.gitmodules
+++ b/.gitmodules
@@ -124,3 +124,9 @@
 [submodule "software/embench/embench-iot"]
 	path = software/embench/embench-iot
 	url = https://github.com/embench/embench-iot.git
+[submodule "shuttle"]
+	path = generators/shuttle
+	url = https://github.com/ucb-bar/shuttle.git
+[submodule "generators/bar-fetchers"]
+	path = generators/bar-fetchers
+	url = https://github.com/ucb-bar/bar-fetchers.git
\ No newline at end of file
diff --git a/build.sbt b/build.sbt
index 6f72940a..64885781 100644
--- a/build.sbt
+++ b/build.sbt
@@ -153,7 +153,7 @@ lazy val chipyard = (project in file("generators/chipyard"))
     sha3, // On separate line to allow for cleaner tutorial-setup patches
     dsptools, `rocket-dsp-utils`,
     gemmini, icenet, tracegen, cva6, nvdla, sodor, ibex, fft_generator,
-    constellation, mempress)
+    constellation, mempress, barf, shuttle)
   .settings(libraryDependencies ++= rocketLibDeps.value)
   .settings(
     libraryDependencies ++= Seq(
@@ -168,6 +168,11 @@ lazy val mempress = (project in file("generators/mempress"))
   .settings(chiselTestSettings)
   .settings(commonSettings)
 
+lazy val barf = (project in file("generators/bar-fetchers"))
+  .dependsOn(rocketchip)
+  .settings(libraryDependencies ++= rocketLibDeps.value)
+  .settings(commonSettings)
+
 lazy val constellation = (project in file("generators/constellation"))
   .dependsOn(rocketchip)
   .settings(libraryDependencies ++= rocketLibDeps.value)
@@ -198,6 +203,11 @@ lazy val boom = (project in file("generators/boom"))
   .settings(libraryDependencies ++= rocketLibDeps.value)
   .settings(commonSettings)
 
+lazy val shuttle = (project in file("generators/shuttle"))
+  .dependsOn(rocketchip)
+  .settings(libraryDependencies ++= rocketLibDeps.value)
+  .settings(commonSettings)
+
 lazy val cva6 = (project in file("generators/cva6"))
   .dependsOn(rocketchip)
   .settings(libraryDependencies ++= rocketLibDeps.value)
diff --git a/common.mk b/common.mk
index 9bb2f94c..7e0f664b 100644
--- a/common.mk
+++ b/common.mk
@@ -68,7 +68,7 @@ include $(base_dir)/tools/torture.mk
 #########################################################################################
 # Prerequisite lists
 #########################################################################################
-# Returns a list of files in directory $1 with file extension $2.
+# Returns a list of files in directories $1 with single file extension $2.
 # If available, use 'fd' to find the list of files, which is faster than 'find'.
 ifeq ($(shell which fd 2> /dev/null),)
 	lookup_srcs = $(shell find -L $(1)/ -name target -prune -o \( -iname "*.$(2)" ! -iname ".*" \) -print 2> /dev/null)
@@ -76,9 +76,17 @@ else
 	lookup_srcs = $(shell fd -L -t f -e $(2) . $(1))
 endif
 
-SOURCE_DIRS = $(addprefix $(base_dir)/,generators sims/firesim/sim tools/barstools fpga/fpga-shells fpga/src)
-SCALA_SOURCES = $(call lookup_srcs,$(SOURCE_DIRS),scala)
-VLOG_SOURCES = $(call lookup_srcs,$(SOURCE_DIRS),sv) $(call lookup_srcs,$(SOURCE_DIRS),v)
+# Returns a list of files in directories $1 with *any* of the file extensions in $2
+lookup_srcs_by_multiple_type = $(foreach type,$(2),$(call lookup_srcs,$(1),$(type)))
+
+SCALA_EXT = scala
+VLOG_EXT = sv v
+CHIPYARD_SOURCE_DIRS = $(addprefix $(base_dir)/,generators sims/firesim/sim fpga/fpga-shells fpga/src)
+CHIPYARD_SCALA_SOURCES = $(call lookup_srcs_by_multiple_type,$(CHIPYARD_SOURCE_DIRS),$(SCALA_EXT))
+CHIPYARD_VLOG_SOURCES = $(call lookup_srcs_by_multiple_type,$(CHIPYARD_SOURCE_DIRS),$(VLOG_EXT))
+BARSTOOLS_SOURCE_DIRS = $(addprefix $(base_dir)/,tools/barstools)
+BARSTOOLS_SCALA_SOURCES = $(call lookup_srcs_by_multiple_type,$(BARSTOOLS_SOURCE_DIRS),$(SCALA_EXT))
+BARSTOOLS_VLOG_SOURCES = $(call lookup_srcs_by_multiple_type,$(BARSTOOLS_SOURCE_DIRS),$(VLOG_EXT))
 # This assumes no SBT meta-build sources
 SBT_SOURCE_DIRS = $(addprefix $(base_dir)/,generators sims/firesim/sim tools)
 SBT_SOURCES = $(call lookup_srcs,$(SBT_SOURCE_DIRS),sbt) $(base_dir)/build.sbt $(base_dir)/project/plugins.sbt $(base_dir)/project/build.properties
@@ -106,12 +114,12 @@ $(BOOTROM_TARGETS): $(build_dir)/bootrom.%.img: $(TESTCHIP_RSRCS_DIR)/testchipip
 #########################################################################################
 # compile scala jars
 #########################################################################################
-$(CHIPYARD_CLASSPATH_TARGETS) &: $(SCALA_SOURCES) $(SCALA_BUILDTOOL_DEPS)
+$(CHIPYARD_CLASSPATH_TARGETS) &: $(CHIPYARD_SCALA_SOURCES) $(SCALA_BUILDTOOL_DEPS)
 	mkdir -p $(dir $@)
 	$(call run_sbt_assembly,$(SBT_PROJECT),$(CHIPYARD_CLASSPATH))
 
 # order only dependency between sbt runs needed to avoid concurrent sbt runs
-$(TAPEOUT_CLASSPATH_TARGETS) &: $(SCALA_SOURCES) $(SCALA_BUILDTOOL_DEPS) | $(CHIPYARD_CLASSPATH_TARGETS)
+$(TAPEOUT_CLASSPATH_TARGETS) &: $(BARSTOOLS_SCALA_SOURCES) $(SCALA_BUILDTOOL_DEPS) | $(CHIPYARD_CLASSPATH_TARGETS)
 	mkdir -p $(dir $@)
 	$(call run_sbt_assembly,tapeout,$(TAPEOUT_CLASSPATH))
 
@@ -200,19 +208,20 @@ else
 endif
 
 $(MFC_LOWERING_OPTIONS):
+	mkdir -p $(dir $@)
 ifeq (,$(ENABLE_YOSYS_FLOW))
 	echo "$(MFC_BASE_LOWERING_OPTIONS)" > $@
 else
 	echo "$(MFC_BASE_LOWERING_OPTIONS),disallowPackedArrays" > $@
 endif
 
-$(FINAL_ANNO_FILE): $(EXTRA_ANNO_FILE) $(SFC_EXTRA_ANNO_FILE) $(VLOG_SOURCES) $(SFC_LEVEL)
+$(FINAL_ANNO_FILE): $(EXTRA_ANNO_FILE) $(SFC_EXTRA_ANNO_FILE) $(SFC_LEVEL)
 	if [ $(shell cat $(SFC_LEVEL)) = low ]; then jq -s '[.[][]]' $(EXTRA_ANNO_FILE) $(SFC_EXTRA_ANNO_FILE) > $@; fi
 	if [ $(shell cat $(SFC_LEVEL)) = none ]; then cat $(EXTRA_ANNO_FILE) > $@; fi
 	touch $@
 
 $(SFC_MFC_TARGETS) &: private TMP_DIR := $(shell mktemp -d -t cy-XXXXXXXX)
-$(SFC_MFC_TARGETS) &: $(TAPEOUT_CLASSPATH_TARGETS) $(FIRRTL_FILE) $(FINAL_ANNO_FILE) $(SFC_LEVEL) $(EXTRA_FIRRTL_OPTIONS) $(MFC_LOWERING_OPTIONS)
+$(SFC_MFC_TARGETS) &: $(TAPEOUT_CLASSPATH_TARGETS) $(FIRRTL_FILE) $(FINAL_ANNO_FILE) $(SFC_LEVEL) $(EXTRA_FIRRTL_OPTIONS) $(MFC_LOWERING_OPTIONS) $(CHIPYARD_VLOG_SOURCES) $(BARSTOOLS_VLOG_SOURCES)
 	rm -rf $(GEN_COLLATERAL_DIR)
 	$(call run_jar_scala_main,$(TAPEOUT_CLASSPATH),barstools.tapeout.transforms.GenerateModelStageMain,\
 		--no-dedup \
diff --git a/docs/Customization/RoCC-Accelerators.rst b/docs/Customization/RoCC-Accelerators.rst
index 79d9e898..ad3bfa54 100644
--- a/docs/Customization/RoCC-Accelerators.rst
+++ b/docs/Customization/RoCC-Accelerators.rst
@@ -1,10 +1,15 @@
 .. _rocc-accelerators:
 
 Adding a RoCC Accelerator
-----------------------------
+-------------------------
 
-RoCC accelerators are lazy modules that extend the ``LazyRoCC`` class.
-Their implementation should extends the ``LazyRoCCModule`` class.
+A RoCC accelerator is a component that can be added into a particular Rocket or BooM tile.
+It receives instructions that match a certain opcode, talks to other parts of the core or SoC (L1, L2, PTW, FPU), and then optionally writes back a value into the register corresponding with the ``rd`` field of the instruction.
+RoCC accelerators are instantiated via modules that extend the ``LazyRoCC`` class.
+These modules lazily instantiate another module which extends the ``LazyRoCCModule`` class.
+This extra layer of indirection is used so that Diplomacy can figure out how to connect the RoCC module to the chip, without needing to instantiate the module ahead of time.
+Lazy modules are further explained in the :ref:`Chipyard-Basics/Configs-Parameters-Mixins:Cake Pattern / Mixin` section.
+Below is a minimal instantiation of a RoCC accelerator.
 
 .. code-block:: scala
 
@@ -31,7 +36,6 @@ Their implementation should extends the ``LazyRoCCModule`` class.
       ...
     }
 
-
 The ``opcodes`` parameter for ``LazyRoCC`` is the set of custom opcodes that will map to this accelerator.
 More on this in the next subsection.
 
@@ -46,6 +50,47 @@ the ``busy`` signal, which indicates when the accelerator is still handling an i
 and the ``interrupt`` signal, which can be used to interrupt the CPU.
 
 Look at the examples in ``generators/rocket-chip/src/main/scala/tile/LazyRoCC.scala`` for detailed information on the different IOs.
+There is also more information about each of the signals in `the RoCC Documentation written by UCSD <https://docs.google.com/document/d/1CH2ep4YcL_ojsa3BVHEW-uwcKh1FlFTjH_kg5v8bxVw/edit>`_, although it is updated out of tree and may be out of date.
+
+
+Accessing Memory via L1 Cache
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+A RoCC accelerator can access memory through the L1 Cache of the core it is attached to.
+This is a simpler interface for accelerator architects to implement, but will generally have lower achievable throughput than a dedicated TileLink port.
+
+In your ``LazyRoCCModuleImp``, the signal ``io.mem`` is a ``HellaCacheIO``, which is defined in ``generators/rocket-chip/src/main/scala/rocket/HellaCache.scala``.
+
+.. code-block:: scala
+
+    class HellaCacheIO(implicit p: Parameters) extends CoreBundle()(p) {
+        val req = Decoupled(new HellaCacheReq)
+        val s1_kill = Output(Bool()) // kill previous cycle's req
+        val s1_data = Output(new HellaCacheWriteData()) // data for previous cycle's req
+        val s2_nack = Input(Bool()) // req from two cycles ago is rejected
+        val s2_nack_cause_raw = Input(Bool()) // reason for nack is store-load RAW hazard (performance hint)
+        val s2_kill = Output(Bool()) // kill req from two cycles ago
+        val s2_uncached = Input(Bool()) // advisory signal that the access is MMIO
+        val s2_paddr = Input(UInt(paddrBits.W)) // translated address
+
+        val resp = Flipped(Valid(new HellaCacheResp))
+        val replay_next = Input(Bool())
+        val s2_xcpt = Input(new HellaCacheExceptions)
+        val s2_gpa = Input(UInt(vaddrBitsExtended.W))
+        val s2_gpa_is_pte = Input(Bool())
+        val uncached_resp = tileParams.dcache.get.separateUncachedResp.option(Flipped(Decoupled(new HellaCacheResp)))
+        val ordered = Input(Bool())
+        val perf = Input(new HellaCachePerfEvents())
+
+        val keep_clock_enabled = Output(Bool()) // should D$ avoid clock-gating itself?
+        val clock_enabled = Input(Bool()) // is D$ currently being clocked?
+    }
+
+At a high level, you must tag requests that you send across this interface using the ``io.mem.req.tag``, and the tag will be returned to you when the data is ready.
+Responses may come back out of order if you issue multiple requests, so you can use these tags to tell what data came back.
+Note that the number of tag bits is controled by ``dcacheReqTagBits``, which is usually set to 6.
+Using more than 6 bits will cause errors or hangs.
+
 
 Adding RoCC accelerator to Config
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
@@ -67,3 +112,4 @@ For instance, if we wanted to add the previously defined accelerator and route c
       new RocketConfig)
 
 To add RoCC instructions in your program, use the RoCC C macros provided in ``tests/rocc.h``. You can find examples in the files ``tests/accum.c`` and ``charcount.c``.
+
diff --git a/docs/Generators/Prefetchers.rst b/docs/Generators/Prefetchers.rst
new file mode 100644
index 00000000..550d22e7
--- /dev/null
+++ b/docs/Generators/Prefetchers.rst
@@ -0,0 +1,9 @@
+Prefetchers
+====================================
+
+The BAR-fetchers library is a collection of Chisel-implemented prefetchers, designed for compatibility with Chipyard and Rocket-Chip SoCs.
+This package implements a generic prefetcher API, and example implementations of NextLine, Strided, and AMPM prefetchers.
+
+Prefetchers can be instantiated in front of a L1D HellaCache, or as TileLink nodes in front of some TileLink bus.
+
+An example configuration using prefetchers is found in the ``PrefetchingRocketConfig``
diff --git a/docs/Generators/Shuttle.rst b/docs/Generators/Shuttle.rst
new file mode 100644
index 00000000..87c8f78b
--- /dev/null
+++ b/docs/Generators/Shuttle.rst
@@ -0,0 +1,8 @@
+Shuttle RISC-V Core
+===================
+
+Shuttle is a Rocket-based superscalar in-order RISC-V core, supporting the base RV64IMAFDC instruction set with supervisor and user-mode. Shuttle is a 6-stage core that can be configured to be dual, three, or quad-issue, although dual-issue is the most sensible design point. Shuttle is not designed to meet any power, performance, or area targets. It exists purely as a demonstrative example of another RISC-V CPU design point.
+
+The superscalar microarchitecture presents the most advantages for 1) floating-point kernels and 2) RoCC accelerator kernels, as scalar control code can execute concurrently with floating point or RoCC instructions, maintaining high utilization of those units.
+
+Shuttle is tape-out proven, and has similar physical design complexity as Rocket.
diff --git a/docs/Generators/index.rst b/docs/Generators/index.rst
index cb8cdc47..c2bf3719 100644
--- a/docs/Generators/index.rst
+++ b/docs/Generators/index.rst
@@ -33,4 +33,6 @@ so changes to the generators themselves will automatically be used when building
    fft
    NVDLA
    Sodor
+   Shuttle
    Mempress
+   Prefetchers
diff --git a/generators/bar-fetchers b/generators/bar-fetchers
new file mode 160000
index 00000000..3a33d818
--- /dev/null
+++ b/generators/bar-fetchers
@@ -0,0 +1 @@
+Subproject commit 3a33d818aefe5444aa27fc1557008f747538d2cc
diff --git a/generators/chipyard/src/main/resources/csrc/cospike.cc b/generators/chipyard/src/main/resources/csrc/cospike.cc
index e1558ba1..bc30021a 100644
--- a/generators/chipyard/src/main/resources/csrc/cospike.cc
+++ b/generators/chipyard/src/main/resources/csrc/cospike.cc
@@ -2,6 +2,8 @@
 #include <vector>
 #include <string>
 #include <riscv/sim.h>
+#include <riscv/mmu.h>
+#include <riscv/encoding.h>
 #include <vpi_user.h>
 #include <svdpi.h>
 #include <sstream>
@@ -27,7 +29,11 @@ extern std::map<long long int, backing_data_t> backing_mem_data;
 #endif
 
 #define CLINT_BASE (0x2000000)
-#define CLINT_SIZE (0x1000)
+#define CLINT_SIZE (0x10000)
+#define UART_BASE (0x54000000)
+#define UART_SIZE (0x1000)
+#define PLIC_BASE (0xc000000)
+#define PLIC_SIZE (0x4000000)
 
 typedef struct system_info_t {
   std::string isa;
@@ -38,13 +44,33 @@ typedef struct system_info_t {
   std::vector<char> bootrom;
 };
 
+class read_override_device_t : public abstract_device_t {
+public:
+  read_override_device_t(std::string n, reg_t sz) : was_read_from(false), size(size), name(n) { };
+  bool load(reg_t addr, size_t len, uint8_t* bytes) {
+    if (addr + len < addr || addr + len > size) return false;
+    printf("Read from device %s at %lx\n", name.c_str(), addr);
+    was_read_from = true;
+    return true;
+  }
+  bool store(reg_t addr, size_t len, const uint8_t* bytes) {
+    return (addr + len >= addr && addr + len <= size);
+  }
+  bool was_read_from;
+private:
+  reg_t size;
+  std::string name;
+};
+
 system_info_t* info = NULL;
 sim_t* sim = NULL;
 bool cospike_debug;
 reg_t tohost_addr = 0;
 reg_t fromhost_addr = 0;
+reg_t cospike_timeout = 0;
 std::set<reg_t> magic_addrs;
 cfg_t* cfg;
+std::vector<read_override_device_t*> read_override_devices;
 
 static std::vector<std::pair<reg_t, mem_t*>> make_mems(const std::vector<mem_cfg_t> &layout)
 {
@@ -90,6 +116,7 @@ extern "C" void cospike_cosim(long long int cycle,
                               int priv)
 {
   assert(info);
+
   if (unlikely(!sim)) {
     printf("Configuring spike cosim\n");
     std::vector<mem_cfg_t> mem_cfg;
@@ -114,19 +141,31 @@ extern "C" void cospike_cosim(long long int cycle,
 
     std::vector<std::pair<reg_t, mem_t*>> mems = make_mems(cfg->mem_layout());
 
+    size_t default_boot_rom_size = 0x10000;
+    size_t default_boot_rom_addr = 0x10000;
+    assert(info->bootrom.size() < default_boot_rom_size);
+    info->bootrom.resize(default_boot_rom_size);
+
     rom_device_t *boot_rom = new rom_device_t(info->bootrom);
     mem_t *boot_addr_reg = new mem_t(0x1000);
     uint64_t default_boot_addr = 0x80000000;
     boot_addr_reg->store(0, 8, (const uint8_t*)(&default_boot_addr));
 
-    // Don't actually build a clint
-    mem_t* clint_mem = new mem_t(CLINT_SIZE);
+    read_override_device_t* clint = new read_override_device_t("clint", CLINT_SIZE);
+    read_override_device_t* uart = new read_override_device_t("uart", UART_SIZE);
+    read_override_device_t* plic = new read_override_device_t("plic", PLIC_SIZE);
+
+    read_override_devices.push_back(clint);
+    read_override_devices.push_back(uart);
+    read_override_devices.push_back(plic);
 
     std::vector<std::pair<reg_t, abstract_device_t*>> plugin_devices;
     // The device map is hardcoded here for now
     plugin_devices.push_back(std::pair(0x4000, boot_addr_reg));
-    plugin_devices.push_back(std::pair(0x10000, boot_rom));
-    plugin_devices.push_back(std::pair(CLINT_BASE, clint_mem));
+    plugin_devices.push_back(std::pair(default_boot_rom_addr, boot_rom));
+    plugin_devices.push_back(std::pair(CLINT_BASE, clint));
+    plugin_devices.push_back(std::pair(UART_BASE, uart));
+    plugin_devices.push_back(std::pair(PLIC_BASE, plic));
 
     s_vpi_vlog_info vinfo;
     if (!vpi_get_vlog_info(&vinfo))
@@ -142,6 +181,8 @@ extern "C" void cospike_cosim(long long int cycle,
         in_permissive = false;
       } else if (arg == "+cospike_debug" || arg == "+cospike-debug") {
         cospike_debug = true;
+      } else if (arg.find("+cospike-timeout=") == 0) {
+	cospike_timeout = strtoull(arg.substr(17).c_str(), 0, 10);
       } else if (!in_permissive) {
         htif_args.push_back(arg);
       }
@@ -159,17 +200,19 @@ extern "C" void cospike_cosim(long long int cycle,
       .support_impebreak = true
     };
 
-    printf("isa string is %s\n", info->isa.c_str());
+    printf("isa string: %s\n", info->isa.c_str());
+    printf("htif args: ");
     for (int i = 0; i < htif_args.size(); i++) {
-      printf("%s\n", htif_args[i].c_str());
+      printf("%s", htif_args[i].c_str());
     }
+    printf("\n");
 
     sim = new sim_t(cfg, false,
                     mems,
                     plugin_devices,
                     htif_args,
                     dm_config,
-                    "cospike.log",
+                    nullptr,
                     false,
                     nullptr,
                     false,
@@ -193,11 +236,19 @@ extern "C" void cospike_cosim(long long int cycle,
 #endif
 
     sim->configure_log(true, true);
-    // Use our own reset vector
     for (int i = 0; i < info->nharts; i++) {
+      // Use our own reset vector
       sim->get_core(hartid)->get_state()->pc = 0x10040;
+      // Set MMU to support up to sv39, as our normal hw configs do
+      sim->get_core(hartid)->set_impl(IMPL_MMU_SV48, false);
+      sim->get_core(hartid)->set_impl(IMPL_MMU_SV57, false);
+
+      // HACKS: Our processor's don't implement zicntr fully, they don't provide time
+      sim->get_core(hartid)->get_state()->csrmap.erase(CSR_TIME);
     }
     sim->set_debug(cospike_debug);
+    sim->set_histogram(true);
+    sim->set_procs_debug(cospike_debug);
     printf("Setting up htif for spike cosim\n");
     ((htif_t*)sim)->start();
     printf("Spike cosim started\n");
@@ -205,14 +256,25 @@ extern "C" void cospike_cosim(long long int cycle,
     fromhost_addr = ((htif_t*)sim)->get_fromhost_addr();
     printf("Tohost  : %lx\n", tohost_addr);
     printf("Fromhost: %lx\n", fromhost_addr);
-    printf("Memory base  : %lx\n", info->mem0_base);
-    printf("Memory Size  : %lx\n", info->mem0_size);
+    printf("BootROM base  : %lx\n", default_boot_rom_addr);
+    printf("BootROM size  : %lx\n", boot_rom->contents().size());
+    printf("Memory  base  : %lx\n", info->mem0_base);
+    printf("Memory  size  : %lx\n", info->mem0_size);
   }
 
   if (priv & 0x4) { // debug
     return;
   }
 
+  if (cospike_timeout && cycle > cospike_timeout) {
+    if (sim) {
+      printf("Cospike reached timeout cycles = %ld, terminating\n", cospike_timeout);
+      delete sim;
+    }
+    exit(0);
+  }
+
+
   processor_t* p = sim->get_core(hartid);
   state_t* s = p->get_state();
 #ifdef COSPIKE_DTM
@@ -269,14 +331,18 @@ extern "C" void cospike_cosim(long long int cycle,
   uint64_t interrupt_cause = cause & 0x7FFFFFFFFFFFFFFF;
   bool ssip_interrupt = interrupt_cause == 0x1;
   bool msip_interrupt = interrupt_cause == 0x3;
+  bool stip_interrupt = interrupt_cause == 0x5;
+  bool mtip_interrupt = interrupt_cause == 0x7;
   bool debug_interrupt = interrupt_cause == 0xe;
   if (raise_interrupt) {
     printf("%d interrupt %lx\n", cycle, cause);
 
-    if (ssip_interrupt) {
+    if (ssip_interrupt || stip_interrupt) {
       // do nothing
     } else if (msip_interrupt) {
       s->mip->backdoor_write_with_mask(MIP_MSIP, MIP_MSIP);
+    } else if (mtip_interrupt) {
+      s->mip->backdoor_write_with_mask(MIP_MTIP, MIP_MTIP);
     } else if (debug_interrupt) {
       return;
     } else {
@@ -295,6 +361,8 @@ extern "C" void cospike_cosim(long long int cycle,
     printf("\n");
   }
   if (valid || raise_interrupt || raise_exception) {
+    p->clear_waiting_for_interrupt();
+    for (auto& e : read_override_devices) e->was_read_from = false;
     p->step(1);
     if (unlikely(cospike_debug)) {
       printf("spike pc is %lx\n", s->pc);
@@ -328,10 +396,8 @@ extern "C" void cospike_cosim(long long int cycle,
       if ((waddr == CLINT_BASE + 4*hartid) && w_data == 0) {
         s->mip->backdoor_write_with_mask(MIP_MSIP, 0);
       }
-      // Try to remember magic_mem addrs, and ignore these in the future
-      if ( waddr == tohost_addr && w_data >= info->mem0_base && w_data < (info->mem0_base + info->mem0_size)) {
-        printf("Probable magic mem %lx\n", w_data);
-        magic_addrs.insert(w_data);
+      if ((waddr == CLINT_BASE + 0x4000 + 4*hartid)) {
+        s->mip->backdoor_write_with_mask(MIP_MTIP, 0);
       }
       // Try to remember magic_mem addrs, and ignore these in the future
       if ( waddr == tohost_addr && w_data >= info->mem0_base && w_data < (info->mem0_base + info->mem0_size)) {
@@ -357,13 +423,18 @@ extern "C" void cospike_cosim(long long int cycle,
       // 2 => vec
       // 3 => vec hint
       // 4 => csr
+      bool device_read = false;
+      for (auto& e : read_override_devices) if (e->was_read_from) device_read = true;
 
-      bool ignore_read = (!mem_read.empty() &&
-                          ((magic_addrs.count(mem_read_addr) ||
+      bool lr_read = ((insn & MASK_LR_D) == MATCH_LR_D) || ((insn & MASK_LR_W) == MATCH_LR_W);
+      bool sc_read = ((insn & MASK_SC_D) == MATCH_SC_D) || ((insn & MASK_SC_W) == MATCH_SC_W);
+
+      bool ignore_read = sc_read || (!mem_read.empty() &&
+                          (magic_addrs.count(mem_read_addr) ||
+                           device_read ||
+                           lr_read ||
                            (tohost_addr && mem_read_addr == tohost_addr) ||
-                           (fromhost_addr && mem_read_addr == fromhost_addr) ||
-                           (CLINT_BASE <= mem_read_addr && mem_read_addr < (CLINT_BASE + CLINT_SIZE)))));
-
+                           (fromhost_addr && mem_read_addr == fromhost_addr)));
       // check the type is compliant with writeback first
       if ((type == 0 || type == 1))
         scalar_wb = true;
@@ -379,11 +450,16 @@ extern "C" void cospike_cosim(long long int cycle,
         bool csr_read = (insn & 0x7f) == 0x73;
         if (csr_read)
           printf("CSR read %lx\n", csr_addr);
-        if (csr_read && ((csr_addr == 0xf13) ||                      // mimpid
+        if (csr_read && ((csr_addr == 0x301) ||                      // misa
+                         (csr_addr == 0x306) ||                      // mcounteren
+                         (csr_addr == 0xf13) ||                      // mimpid
                          (csr_addr == 0xf12) ||                      // marchid
                          (csr_addr == 0xf11) ||                      // mvendorid
                          (csr_addr == 0xb00) ||                      // mcycle
                          (csr_addr == 0xb02) ||                      // minstret
+                         (csr_addr == 0xc00) ||                      // cycle
+                         (csr_addr == 0xc01) ||                      // time
+                         (csr_addr == 0xc02) ||                      // instret
                          (csr_addr >= 0x7a0 && csr_addr <= 0x7aa) || // debug trigger registers
                          (csr_addr >= 0x3b0 && csr_addr <= 0x3ef)    // pmpaddr
                          )) {
@@ -394,7 +470,7 @@ extern "C" void cospike_cosim(long long int cycle,
           // from clint Technically this could be buggy because log_mem_read
           // only reports vaddrs, but no software ever should access
           // tohost/fromhost/clint with vaddrs anyways
-          printf("Read override %lx\n", mem_read_addr);
+          printf("Read override %lx = %lx\n", mem_read_addr, wdata);
           s->XPR.write(rd, wdata);
         } else if (wdata != regwrite.second.v[0]) {
           printf("%d wdata mismatch reg %d %lx != %lx\n", cycle, rd,
diff --git a/generators/chipyard/src/main/scala/DigitalTop.scala b/generators/chipyard/src/main/scala/DigitalTop.scala
index 6711eac7..561af0d7 100644
--- a/generators/chipyard/src/main/scala/DigitalTop.scala
+++ b/generators/chipyard/src/main/scala/DigitalTop.scala
@@ -16,7 +16,7 @@ class DigitalTop(implicit p: Parameters) extends ChipyardSystem
   with testchipip.CanHavePeripheryCustomBootPin // Enables optional custom boot pin
   with testchipip.CanHavePeripheryBootAddrReg // Use programmable boot address register
   with testchipip.CanHaveTraceIO // Enables optionally adding trace IO
-  with testchipip.CanHaveBackingScratchpad // Enables optionally adding a backing scratchpad
+  with testchipip.CanHaveBankedScratchpad // Enables optionally adding a banked scratchpad
   with testchipip.CanHavePeripheryBlockDevice // Enables optionally adding the block device
   with testchipip.CanHavePeripheryTLSerial // Enables optionally adding the backing memory and serial adapter
   with sifive.blocks.devices.i2c.HasPeripheryI2C // Enables optionally adding the sifive I2C
diff --git a/generators/chipyard/src/main/scala/config/NoCConfigs.scala b/generators/chipyard/src/main/scala/config/NoCConfigs.scala
index 0036e988..0b6be558 100644
--- a/generators/chipyard/src/main/scala/config/NoCConfigs.scala
+++ b/generators/chipyard/src/main/scala/config/NoCConfigs.scala
@@ -68,9 +68,9 @@ class MultiNoCConfig extends Config(
         "serial-tl" -> 0),
       outNodeMapping = ListMap(
         "error" -> 1, "l2[0]" -> 2, "pbus" -> 3, "plic" -> 4,
-        "clint" -> 5, "dmInner" -> 6, "bootrom" -> 7, "tileClockGater" -> 8, "tileResetSetter" -> 9)),
+        "clint" -> 5, "dmInner" -> 6, "bootrom" -> 7, "clock" -> 8)),
     NoCParams(
-      topology = TerminalRouter(BidirectionalLine(10)),
+      topology = TerminalRouter(BidirectionalLine(9)),
       channelParamGen = (a, b) => UserChannelParams(Seq.fill(5) { UserVirtualChannelParams(4) }),
       routingRelation = NonblockingVirtualSubnetworksRouting(TerminalRouterRouting(BidirectionalLineRouting()), 5, 1))
   )) ++
diff --git a/generators/chipyard/src/main/scala/config/RocketConfigs.scala b/generators/chipyard/src/main/scala/config/RocketConfigs.scala
index 21b630cc..d60a58e8 100644
--- a/generators/chipyard/src/main/scala/config/RocketConfigs.scala
+++ b/generators/chipyard/src/main/scala/config/RocketConfigs.scala
@@ -74,13 +74,19 @@ class L1ScratchpadRocketConfig extends Config(
   new chipyard.config.AbstractConfig)
 
 // DOC include start: mbusscratchpadrocket
-class MbusScratchpadRocketConfig extends Config(
-  new testchipip.WithBackingScratchpad ++                   // add mbus backing scratchpad
-  new freechips.rocketchip.subsystem.WithNoMemPort ++       // remove offchip mem port
+class MbusScratchpadOnlyRocketConfig extends Config(
+  new testchipip.WithMbusScratchpad(banks=2, partitions=2) ++               // add 2 partitions of 2 banks mbus backing scratchpad
+  new freechips.rocketchip.subsystem.WithNoMemPort ++         // remove offchip mem port
   new freechips.rocketchip.subsystem.WithNBigCores(1) ++
   new chipyard.config.AbstractConfig)
 // DOC include end: mbusscratchpadrocket
 
+class SbusScratchpadRocketConfig extends Config(
+  new testchipip.WithSbusScratchpad(base=0x70000000L, banks=4) ++ // add 4 banks sbus backing scratchpad
+  new freechips.rocketchip.subsystem.WithNBigCores(1) ++
+  new chipyard.config.AbstractConfig)
+
+
 class MulticlockRocketConfig extends Config(
   new freechips.rocketchip.subsystem.WithAsynchronousRocketTiles(3, 3) ++ // Add async crossings between RocketTile and uncore
   new freechips.rocketchip.subsystem.WithNBigCores(1) ++
@@ -126,3 +132,12 @@ class CustomIOChipTopRocketConfig extends Config(
   new chipyard.example.WithCustomIOCells ++
   new freechips.rocketchip.subsystem.WithNBigCores(1) ++         // single rocket-core
   new chipyard.config.AbstractConfig)
+
+class PrefetchingRocketConfig extends Config(
+  new barf.WithHellaCachePrefetcher(Seq(0), barf.SingleStridedPrefetcherParams()) ++   // strided prefetcher, sits in front of the L1D$, monitors core requests to prefetching into the L1D$
+  new barf.WithTLICachePrefetcher(barf.MultiNextLinePrefetcherParams()) ++             // next-line prefetcher, sits between L1I$ and L2, monitors L1I$ misses to prefetch into L2
+  new barf.WithTLDCachePrefetcher(barf.SingleAMPMPrefetcherParams()) ++                // AMPM prefetcher, sits between L1D$ and L2, monitors L1D$ misses to prefetch into L2
+  new chipyard.config.WithTilePrefetchers ++                                           // add TL prefetchers between tiles and the sbus
+  new freechips.rocketchip.subsystem.WithNonblockingL1(2) ++                           // non-blocking L1D$, L1 prefetching only works with non-blocking L1D$
+  new freechips.rocketchip.subsystem.WithNBigCores(1) ++                               // single rocket-core
+  new chipyard.config.AbstractConfig)
diff --git a/generators/chipyard/src/main/scala/config/ShuttleConfigs.scala b/generators/chipyard/src/main/scala/config/ShuttleConfigs.scala
new file mode 100644
index 00000000..10220bdb
--- /dev/null
+++ b/generators/chipyard/src/main/scala/config/ShuttleConfigs.scala
@@ -0,0 +1,32 @@
+package chipyard
+
+import org.chipsalliance.cde.config.{Config}
+
+//-----------------
+// Shuttle Configs
+//-----------------
+
+class ShuttleConfig extends Config(
+  new shuttle.common.WithNShuttleCores ++                        // 1x dual-issue shuttle core
+  new chipyard.config.AbstractConfig)
+
+class ShuttleCosimConfig extends Config(
+  new chipyard.harness.WithCospike ++                            // attach spike-cosim
+  new chipyard.config.WithTraceIO ++                             // enable trace-io for cosim
+  new shuttle.common.WithShuttleDebugROB ++                      // enable shuttle debug ROB for cosim
+  new shuttle.common.WithNShuttleCores ++
+  new chipyard.config.AbstractConfig)
+
+class dmiShuttleCosimConfig extends Config(
+  new chipyard.harness.WithSerialTLTiedOff ++                    // don't attach anything to serial-tl
+  new chipyard.harness.WithCospike ++                            // attach spike-cosim
+  new chipyard.config.WithDMIDTM ++                              // have debug module expose a clocked DMI port
+  new chipyard.config.WithTraceIO ++                             // enable traceio for cosim
+  new shuttle.common.WithShuttleDebugROB ++                      // enable shuttle debug ROB for cosim
+  new shuttle.common.WithNShuttleCores ++
+  new chipyard.config.AbstractConfig)
+
+class GemminiShuttleConfig extends Config(
+  new gemmini.DefaultGemminiConfig ++                            // use Gemmini systolic array GEMM accel
+  new shuttle.common.WithNShuttleCores ++
+  new chipyard.config.AbstractConfig)
diff --git a/generators/chipyard/src/main/scala/config/fragments/PeripheralFragments.scala b/generators/chipyard/src/main/scala/config/fragments/PeripheralFragments.scala
index 3b607ae0..09636c49 100644
--- a/generators/chipyard/src/main/scala/config/fragments/PeripheralFragments.scala
+++ b/generators/chipyard/src/main/scala/config/fragments/PeripheralFragments.scala
@@ -6,7 +6,8 @@ import chisel3.util.{log2Up}
 
 import org.chipsalliance.cde.config.{Config}
 import freechips.rocketchip.devices.tilelink.{BootROMLocated, PLICKey}
-import freechips.rocketchip.devices.debug.{Debug, ExportDebug, DebugModuleKey, DMI}
+import freechips.rocketchip.devices.debug.{Debug, ExportDebug, DebugModuleKey, DMI, JtagDTMKey, JtagDTMConfig}
+import freechips.rocketchip.diplomacy.{AsynchronousCrossing}
 import freechips.rocketchip.stage.phases.TargetDirKey
 import freechips.rocketchip.subsystem._
 import freechips.rocketchip.tile.{XLen}
@@ -14,49 +15,121 @@ import freechips.rocketchip.tile.{XLen}
 import sifive.blocks.devices.gpio._
 import sifive.blocks.devices.uart._
 import sifive.blocks.devices.spi._
+import sifive.blocks.devices.i2c._
 
 import testchipip._
 
 import chipyard.{ExtTLMem}
 
-// Set the bootrom to the Chipyard bootrom
-class WithBootROM extends Config((site, here, up) => {
+/**
+  * Config fragment for adding a BootROM to the SoC
+  * 
+  * @param address the address of the BootROM device
+  * @param size the size of the BootROM
+  * @param hang the power-on reset vector, i.e. the program counter will be set to this value on reset
+  * @param contentFileName the path to the BootROM image
+  */
+class WithBootROM(address: BigInt = 0x10000, size: Int = 0x10000, hang: BigInt = 0x10040) extends Config((site, here, up) => {
   case BootROMLocated(x) => up(BootROMLocated(x), site)
-      .map(_.copy(contentFileName = s"${site(TargetDirKey)}/bootrom.rv${site(XLen)}.img"))
+      .map(_.copy(
+        address = address,
+        size = size,
+        hang = hang,
+        contentFileName = s"${site(TargetDirKey)}/bootrom.rv${site(XLen)}.img"
+      ))
 })
 
 // DOC include start: gpio config fragment
-class WithGPIO extends Config((site, here, up) => {
-  case PeripheryGPIOKey => Seq(
-    GPIOParams(address = 0x10012000, width = 4, includeIOF = false))
+/**
+ * Config fragment for adding a GPIO peripheral device to the SoC
+ * 
+ * @param address the address of the GPIO device
+ * @param width the number of pins of the GPIO device
+ */
+class WithGPIO(address: BigInt = 0x10010000, width: Int = 4) extends Config ((site, here, up) => {
+  case PeripheryGPIOKey => up(PeripheryGPIOKey) ++ Seq(
+    GPIOParams(address = address, width = width, includeIOF = false))
 })
 // DOC include end: gpio config fragment
 
-class WithUART(baudrate: BigInt = 115200) extends Config((site, here, up) => {
-  case PeripheryUARTKey => Seq(
-    UARTParams(address = 0x54000000L, nTxEntries = 256, nRxEntries = 256, initBaudRate = baudrate))
-})
-
+/**
+ * Config fragment for removing all UART peripheral devices from the SoC
+ */
 class WithNoUART extends Config((site, here, up) => {
   case PeripheryUARTKey => Nil
 })
 
+/**
+  * Config fragment for adding a UART peripheral device to the SoC
+  *
+  * @param address the address of the UART device
+  * @param baudrate the baudrate of the UART device
+  */
+class WithUART(address: BigInt = 0x10020000, baudrate: BigInt = 115200) extends Config ((site, here, up) => {
+  case PeripheryUARTKey => up(PeripheryUARTKey) ++ Seq(
+    UARTParams(address = address, nTxEntries = 256, nRxEntries = 256, initBaudRate = baudrate))
+})
+
 class WithUARTFIFOEntries(txEntries: Int, rxEntries: Int) extends Config((site, here, up) => {
   case PeripheryUARTKey => up(PeripheryUARTKey).map(_.copy(nTxEntries = txEntries, nRxEntries = rxEntries))
 })
 
-class WithSPIFlash(size: BigInt = 0x10000000) extends Config((site, here, up) => {
+/**
+  * Config fragment for adding a SPI peripheral device with Execute-in-Place capability to the SoC
+  *
+  * @param address the address of the SPI controller
+  * @param fAddress the address of the Execute-in-Place (XIP) region of the SPI flash memory
+  * @param size the size of the Execute-in-Place (XIP) region of the SPI flash memory
+  */
+class WithSPIFlash(address: BigInt = 0x10030000, fAddress: BigInt = 0x20000000, size: BigInt = 0x10000000) extends Config((site, here, up) => {
   // Note: the default size matches freedom with the addresses below
-  case PeripherySPIFlashKey => Seq(
-    SPIFlashParams(rAddress = 0x10040000, fAddress = 0x20000000, fSize = size))
+  case PeripherySPIFlashKey => up(PeripherySPIFlashKey) ++ Seq(
+    SPIFlashParams(rAddress = address, fAddress = fAddress, fSize = size))
+})
+
+/**
+  * Config fragment for adding a SPI peripheral device to the SoC
+  *
+  * @param address the address of the SPI controller
+  */
+class WithSPI(address: BigInt = 0x10031000) extends Config((site, here, up) => {
+  case PeripherySPIKey => up(PeripherySPIKey) ++ Seq(
+    SPIParams(rAddress = address))
+})
+
+/**
+  * Config fragment for adding a I2C peripheral device to the SoC
+  *
+  * @param address the address of the I2C controller
+  */
+class WithI2C(address: BigInt = 0x10040000) extends Config((site, here, up) => {
+  case PeripheryI2CKey => up(PeripheryI2CKey) ++ Seq(
+    I2CParams(address = address, controlXType = AsynchronousCrossing(), intXType = AsynchronousCrossing())
+  )
+})
+
+class WithNoDebug extends Config((site, here, up) => {
+  case DebugModuleKey => None
 })
 
 class WithDMIDTM extends Config((site, here, up) => {
   case ExportDebug => up(ExportDebug, site).copy(protocols = Set(DMI))
 })
 
-class WithNoDebug extends Config((site, here, up) => {
-  case DebugModuleKey => None
+/**
+  * Config fragment for adding a JTAG Debug Module to the SoC
+  *
+  * @param idcodeVersion the version of the JTAG protocol the Debug Module supports
+  * @param partNum the part number of the Debug Module
+  * @param manufId the 11-bit JEDEC Designer ID of the chip manufacturer
+  * @param debugIdleCycles the number of cycles the Debug Module waits before responding to a request
+  */
+class WithJTAGDTMKey(idcodeVersion: Int = 2, partNum: Int = 0x000, manufId: Int = 0x489, debugIdleCycles: Int = 5) extends Config((site, here, up) => {
+  case JtagDTMKey => new JtagDTMConfig (
+    idcodeVersion = idcodeVersion,
+    idcodePartNum = partNum,
+    idcodeManufId = manufId,
+    debugIdleCycles = debugIdleCycles)
 })
 
 class WithTLBackingMemory extends Config((site, here, up) => {
diff --git a/generators/chipyard/src/main/scala/config/fragments/TileFragments.scala b/generators/chipyard/src/main/scala/config/fragments/TileFragments.scala
index 56042c3d..17eaa3f0 100644
--- a/generators/chipyard/src/main/scala/config/fragments/TileFragments.scala
+++ b/generators/chipyard/src/main/scala/config/fragments/TileFragments.scala
@@ -9,7 +9,10 @@ import freechips.rocketchip.rocket.{RocketCoreParams, MulDivParams, DCacheParams
 
 import boom.common.{BoomTileAttachParams}
 import cva6.{CVA6TileAttachParams}
+import sodor.common.{SodorTileAttachParams}
+import ibex.{IbexTileAttachParams}
 import testchipip._
+import barf.{TilePrefetchingMasterPortParams}
 
 class WithL2TLBs(entries: Int) extends Config((site, here, up) => {
   case TilesLocated(InSubsystem) => up(TilesLocated(InSubsystem), site) map {
@@ -79,3 +82,17 @@ class WithRocketDCacheScratchpad extends Config((site, here, up) => {
   }
 })
 
+class WithTilePrefetchers extends Config((site, here, up) => {
+  case TilesLocated(InSubsystem) => up(TilesLocated(InSubsystem), site) map {
+    case tp: RocketTileAttachParams => tp.copy(crossingParams = tp.crossingParams.copy(
+      master = TilePrefetchingMasterPortParams(tp.tileParams.hartId, tp.crossingParams.master)))
+    case tp: BoomTileAttachParams => tp.copy(crossingParams = tp.crossingParams.copy(
+      master = TilePrefetchingMasterPortParams(tp.tileParams.hartId, tp.crossingParams.master)))
+    case tp: SodorTileAttachParams => tp.copy(crossingParams = tp.crossingParams.copy(
+      master = TilePrefetchingMasterPortParams(tp.tileParams.hartId, tp.crossingParams.master)))
+    case tp: IbexTileAttachParams => tp.copy(crossingParams = tp.crossingParams.copy(
+      master = TilePrefetchingMasterPortParams(tp.tileParams.hartId, tp.crossingParams.master)))
+    case tp: CVA6TileAttachParams => tp.copy(crossingParams = tp.crossingParams.copy(
+      master = TilePrefetchingMasterPortParams(tp.tileParams.hartId, tp.crossingParams.master)))
+  }
+})
diff --git a/generators/constellation b/generators/constellation
index e9f1c828..8184e0e7 160000
--- a/generators/constellation
+++ b/generators/constellation
@@ -1 +1 @@
-Subproject commit e9f1c828ca5adb4fa46a242cd1798391fc9e6f62
+Subproject commit 8184e0e7e32ff11dce344c449158aa7551e164e0
diff --git a/generators/firechip/src/main/scala/TargetConfigs.scala b/generators/firechip/src/main/scala/TargetConfigs.scala
index c27abd50..1c821294 100644
--- a/generators/firechip/src/main/scala/TargetConfigs.scala
+++ b/generators/firechip/src/main/scala/TargetConfigs.scala
@@ -138,7 +138,7 @@ class WithFireSimConfigTweaks extends Config(
 class WithMinimalFireSimHighPerfConfigTweaks extends Config(
   new WithFireSimHighPerfClocking ++
   new freechips.rocketchip.subsystem.WithNoMemPort ++
-  new testchipip.WithBackingScratchpad ++
+  new testchipip.WithMbusScratchpad ++
   new WithMinimalFireSimDesignTweaks
 )
 
@@ -148,7 +148,7 @@ class WithMinimalFireSimHighPerfConfigTweaks extends Config(
 class WithMinimalAndBlockDeviceFireSimHighPerfConfigTweaks extends Config(
   new WithFireSimHighPerfClocking ++
   new freechips.rocketchip.subsystem.WithNoMemPort ++ // removes mem port for FASEDBridge to match against
-  new testchipip.WithBackingScratchpad ++ // adds backing scratchpad for memory to replace FASED model
+  new testchipip.WithMbusScratchpad ++ // adds backing scratchpad for memory to replace FASED model
   new testchipip.WithBlockDevice(true) ++ // add in block device
   new WithMinimalFireSimDesignTweaks
 )
@@ -329,7 +329,7 @@ class FireSim16LargeBoomConfig extends Config(
 class FireSimNoMemPortConfig extends Config(
   new WithDefaultFireSimBridges ++
   new freechips.rocketchip.subsystem.WithNoMemPort ++
-  new testchipip.WithBackingScratchpad ++
+  new testchipip.WithMbusScratchpad ++
   new WithFireSimConfigTweaks ++
   new chipyard.RocketConfig)
 
diff --git a/generators/shuttle b/generators/shuttle
new file mode 160000
index 00000000..3c15591a
--- /dev/null
+++ b/generators/shuttle
@@ -0,0 +1 @@
+Subproject commit 3c15591a9ee237bb1251e5362de22ad19f64ba07
diff --git a/generators/testchipip b/generators/testchipip
index a3e9c1ff..35d7e196 160000
--- a/generators/testchipip
+++ b/generators/testchipip
@@ -1 +1 @@
-Subproject commit a3e9c1ffeae8af573831e4ac5fd00a76df0ca7f1
+Subproject commit 35d7e1969d1d3e54d29a10901737d9b2ba2ab5a3
diff --git a/scripts/tutorial-patches/build.sbt.patch b/scripts/tutorial-patches/build.sbt.patch
index db81b052..04274910 100644
--- a/scripts/tutorial-patches/build.sbt.patch
+++ b/scripts/tutorial-patches/build.sbt.patch
@@ -10,7 +10,7 @@ index ec36a85f..c0c2849a 100644
 +//    sha3, // On separate line to allow for cleaner tutorial-setup patches
      dsptools, `rocket-dsp-utils`,
      gemmini, icenet, tracegen, cva6, nvdla, sodor, ibex, fft_generator,
-     constellation, mempress)
+     constellation, mempress, barf, shuttle)
 @@ -204,11 +204,11 @@ lazy val sodor = (project in file("generators/riscv-sodor"))
    .settings(libraryDependencies ++= rocketLibDeps.value)
    .settings(commonSettings)