# modified or unmodified, in source code or in binary form.
#
# Copyright (c) 2013 Amin Farmahini-Farahani
+# Copyright (c) 2015 University of Kaiserslautern
# All rights reserved.
#
# Redistribution and use in source and binary forms, with or without
#
# Authors: Andreas Hansson
# Ani Udipi
+# Omar Naji
+# Matthias Jung
from m5.params import *
from AbstractMemory import *
# bus in front of the controller for multiple ports
port = SlavePort("Slave port")
- # the basic configuration of the controller architecture
+ # the basic configuration of the controller architecture, note
+ # that each entry corresponds to a burst for the specific DRAM
+ # configuration (e.g. x32 with burst length 8 is 32 bytes) and not
+ # the cacheline size or request/packet size
write_buffer_size = Param.Unsigned(64, "Number of write queue entries")
read_buffer_size = Param.Unsigned(32, "Number of read queue entries")
# scheduler, address map and page policy
mem_sched_policy = Param.MemSched('frfcfs', "Memory scheduling policy")
- addr_mapping = Param.AddrMap('RoRaBaChCo', "Address mapping policy")
+ addr_mapping = Param.AddrMap('RoRaBaCoCh', "Address mapping policy")
page_policy = Param.PageManage('open_adaptive', "Page management policy")
# enforce a limit on the number of accesses per row
max_accesses_per_row = Param.Unsigned(16, "Max accesses per row before "
"closing");
+ # size of DRAM Chip in Bytes
+ device_size = Param.MemorySize("Size of DRAM chip")
+
# pipeline latency of the controller and PHY, split into a
# frontend part and a backend part, with reads and writes serviced
# by the queues only seeing the frontend contribution, and reads
"device/chip")
devices_per_rank = Param.Unsigned("Number of devices/chips per rank")
ranks_per_channel = Param.Unsigned("Number of ranks per channel")
+
+ # default to 0 bank groups per rank, indicating bank group architecture
+ # is not used
+ # update per memory class when bank group architecture is supported
+ bank_groups_per_rank = Param.Unsigned(0, "Number of bank groups per rank")
banks_per_rank = Param.Unsigned("Number of banks per rank")
# only used for the address mapping as the controller by
# construction is a single channel and multiple controllers have
# to be instantiated for a multi-channel configuration
channels = Param.Unsigned(1, "Number of channels")
+ # For power modelling we need to know if the DRAM has a DLL or not
+ dll = Param.Bool(True, "DRAM has DLL or not")
+
+ # DRAMPower provides in addition to the core power, the possibility to
+ # include RD/WR termination and IO power. This calculation assumes some
+ # default values. The integration of DRAMPower with gem5 does not include
+ # IO and RD/WR termination power by default. This might be added as an
+ # additional feature in the future.
+
# timing behaviour and constraints - all in nanoseconds
# the base clock period of the DRAM
# This parameter has to account for burst length.
# Read/Write requests with data size larger than one full burst are broken
# down into multiple requests in the controller
+ # tBURST is equivalent to the CAS-to-CAS delay (tCCD)
+ # With bank group architectures, tBURST represents the CAS-to-CAS
+ # delay for bursts to different bank groups (tCCD_S)
tBURST = Param.Latency("Burst duration (for DDR burst length / 2 cycles)")
+ # CAS-to-CAS delay for bursts to the same bank group
+ # only utilized with bank group architectures; set to 0 for default case
+ # tBURST is equivalent to tCCD_S; no explicit parameter required
+ # for CAS-to-CAS delay for bursts to different bank groups
+ tCCD_L = Param.Latency("0ns", "Same bank group CAS to CAS delay")
+
# time taken to complete one refresh cycle (N rows in all banks)
tRFC = Param.Latency("Refresh cycle time")
# to be sent. It is 7.8 us for a 64ms refresh requirement
tREFI = Param.Latency("Refresh command interval")
- # write-to-read turn around penalty
- tWTR = Param.Latency("Write to read switching time")
+ # write-to-read, same rank turnaround penalty
+ tWTR = Param.Latency("Write to read, same rank switching time")
- # read-to-write turn around penalty, bus turnaround delay
- tRTW = Param.Latency("Read to write switching time")
+ # read-to-write, same rank turnaround penalty
+ tRTW = Param.Latency("Read to write, same rank switching time")
+
+ # rank-to-rank bus delay penalty
+ # this does not correlate to a memory timing parameter and encompasses:
+ # 1) RD-to-RD, 2) WR-to-WR, 3) RD-to-WR, and 4) WR-to-RD
+ # different rank bus delay
+ tCS = Param.Latency("Rank to rank switching time")
# minimum row activate to row activate delay time
tRRD = Param.Latency("ACT to ACT delay")
+ # only utilized with bank group architectures; set to 0 for default case
+ tRRD_L = Param.Latency("0ns", "Same bank group ACT to ACT delay")
+
# time window in which a maximum number of activates are allowed
# to take place, set to 0 to disable
tXAW = Param.Latency("X activation window")
activation_limit = Param.Unsigned("Max number of activates in window")
+ # time to exit power-down mode
+ # Exit power-down to next valid command delay
+ tXP = Param.Latency("0ns", "Power-up Delay")
+
+ # Exit Powerdown to commands requiring a locked DLL
+ tXPDLL = Param.Latency("0ns", "Power-up Delay with locked DLL")
+
+ # time to exit self-refresh mode
+ tXS = Param.Latency("0ns", "Self-refresh exit latency")
+
+ # time to exit self-refresh mode with locked DLL
+ tXSDLL = Param.Latency("0ns", "Self-refresh exit latency DLL")
+
# Currently rolled into other params
######################################################################
# tRC - assumed to be tRAS + tRP
-# A single DDR3 x64 interface (one command and address bus), with
-# default timings based on DDR3-1600 4 Gbit parts in an 8x8
-# configuration, which would amount to 4 Gbyte of memory.
+ # Power Behaviour and Constraints
+ # DRAMs like LPDDR and WideIO have 2 external voltage domains. These are
+ # defined as VDD and VDD2. Each current is defined for each voltage domain
+ # separately. For example, current IDD0 is active-precharge current for
+ # voltage domain VDD and current IDD02 is active-precharge current for
+ # voltage domain VDD2.
+ # By default all currents are set to 0mA. Users who are only interested in
+ # the performance of DRAMs can leave them at 0.
+
+ # Operating 1 Bank Active-Precharge current
+ IDD0 = Param.Current("0mA", "Active precharge current")
+
+ # Operating 1 Bank Active-Precharge current multiple voltage Range
+ IDD02 = Param.Current("0mA", "Active precharge current VDD2")
+
+ # Precharge Power-down Current: Slow exit
+ IDD2P0 = Param.Current("0mA", "Precharge Powerdown slow")
+
+ # Precharge Power-down Current: Slow exit multiple voltage Range
+ IDD2P02 = Param.Current("0mA", "Precharge Powerdown slow VDD2")
+
+ # Precharge Power-down Current: Fast exit
+ IDD2P1 = Param.Current("0mA", "Precharge Powerdown fast")
+
+ # Precharge Power-down Current: Fast exit multiple voltage Range
+ IDD2P12 = Param.Current("0mA", "Precharge Powerdown fast VDD2")
+
+ # Precharge Standby current
+ IDD2N = Param.Current("0mA", "Precharge Standby current")
+
+ # Precharge Standby current multiple voltage range
+ IDD2N2 = Param.Current("0mA", "Precharge Standby current VDD2")
+
+ # Active Power-down current: slow exit
+ IDD3P0 = Param.Current("0mA", "Active Powerdown slow")
+
+ # Active Power-down current: slow exit multiple voltage range
+ IDD3P02 = Param.Current("0mA", "Active Powerdown slow VDD2")
+
+ # Active Power-down current : fast exit
+ IDD3P1 = Param.Current("0mA", "Active Powerdown fast")
+
+ # Active Power-down current : fast exit multiple voltage range
+ IDD3P12 = Param.Current("0mA", "Active Powerdown fast VDD2")
+
+ # Active Standby current
+ IDD3N = Param.Current("0mA", "Active Standby current")
+
+ # Active Standby current multiple voltage range
+ IDD3N2 = Param.Current("0mA", "Active Standby current VDD2")
+
+ # Burst Read Operating Current
+ IDD4R = Param.Current("0mA", "READ current")
+
+ # Burst Read Operating Current multiple voltage range
+ IDD4R2 = Param.Current("0mA", "READ current VDD2")
+
+ # Burst Write Operating Current
+ IDD4W = Param.Current("0mA", "WRITE current")
+
+ # Burst Write Operating Current multiple voltage range
+ IDD4W2 = Param.Current("0mA", "WRITE current VDD2")
+
+ # Refresh Current
+ IDD5 = Param.Current("0mA", "Refresh current")
+
+ # Refresh Current multiple voltage range
+ IDD52 = Param.Current("0mA", "Refresh current VDD2")
+
+ # Self-Refresh Current
+ IDD6 = Param.Current("0mA", "Self-refresh Current")
+
+ # Self-Refresh Current multiple voltage range
+ IDD62 = Param.Current("0mA", "Self-refresh Current VDD2")
+
+ # Main voltage range of the DRAM
+ VDD = Param.Voltage("0V", "Main Voltage Range")
+
+ # Second voltage range defined by some DRAMs
+ VDD2 = Param.Voltage("0V", "2nd Voltage Range")
+
+# A single DDR3-1600 x64 channel (one command and address bus), with
+# timings based on a DDR3-1600 4 Gbit datasheet (Micron MT41J512M8) in
+# an 8x8 configuration.
class DDR3_1600_x64(DRAMCtrl):
+ # size of device in bytes
+ device_size = '512MB'
+
# 8x8 configuration, 8 devices each with an 8-bit interface
device_bus_width = 8
# DDR3 is a BL8 device
burst_length = 8
- # Each device has a page (row buffer) size of 1KB
- # (this depends on the memory density)
+ # Each device has a page (row buffer) size of 1 Kbyte (1K columns x8)
device_rowbuffer_size = '1kB'
# 8x8 configuration, so 8 devices
# 800 MHz
tCK = '1.25ns'
- # DDR3-1600 11-11-11-28
+ # 8 beats across an x64 interface translates to 4 clocks @ 800 MHz
+ tBURST = '5ns'
+
+ # DDR3-1600 11-11-11
tRCD = '13.75ns'
tCL = '13.75ns'
tRP = '13.75ns'
tRAS = '35ns'
+ tRRD = '6ns'
+ tXAW = '30ns'
+ activation_limit = 4
+ tRFC = '260ns'
+
tWR = '15ns'
+
+ # Greater of 4 CK or 7.5 ns
+ tWTR = '7.5ns'
+
+ # Greater of 4 CK or 7.5 ns
tRTP = '7.5ns'
- # 8 beats across an x64 interface translates to 4 clocks @ 800 MHz.
- # Note this is a BL8 DDR device.
- tBURST = '5ns'
+ # Default same rank rd-to-wr bus turnaround to 2 CK, @800 MHz = 2.5 ns
+ tRTW = '2.5ns'
- # DDR3, 4 Gbit has a tRFC of 240 CK and tCK = 1.25 ns
- tRFC = '300ns'
+ # Default different rank bus delay to 2 CK, @800 MHz = 2.5 ns
+ tCS = '2.5ns'
- # DDR3, <=85C, half for >85C
+ # <=85C, half for >85C
tREFI = '7.8us'
- # Greater of 4 CK or 7.5 ns, 4 CK @ 800 MHz = 5 ns
- tWTR = '7.5ns'
+ # Current values from datasheet
+ IDD0 = '75mA'
+ IDD2N = '50mA'
+ IDD3N = '57mA'
+ IDD4W = '165mA'
+ IDD4R = '187mA'
+ IDD5 = '220mA'
+ VDD = '1.5V'
+
+# A single HMC-2500 x32 model based on:
+# [1] DRAMSpec: a high-level DRAM bank modelling tool
+# developed at the University of Kaiserslautern. This high level tool
+# uses RC (resistance-capacitance) and CV (capacitance-voltage) models to
+# estimate the DRAM bank latency and power numbers.
+# [2] A Logic-base Interconnect for Supporting Near Memory Computation in the
+# Hybrid Memory Cube (E. Azarkhish et. al)
+# Assumed for the HMC model is a 30 nm technology node.
+# The modelled HMC consists of 4 Gbit layers which sum up to 2GB of memory (4
+# layers).
+# Each layer has 16 vaults and each vault consists of 2 banks per layer.
+# In order to be able to use the same controller used for 2D DRAM generations
+# for HMC, the following analogy is done:
+# Channel (DDR) => Vault (HMC)
+# device_size (DDR) => size of a single layer in a vault
+# ranks per channel (DDR) => number of layers
+# banks per rank (DDR) => banks per layer
+# devices per rank (DDR) => devices per layer ( 1 for HMC).
+# The parameters for which no input is available are inherited from the DDR3
+# configuration.
+# This configuration includes the latencies from the DRAM to the logic layer of
+# the HMC
+class HMC_2500_x32(DDR3_1600_x64):
+ # size of device
+ # two banks per device with each bank 4MB [2]
+ device_size = '8MB'
+
+ # 1x32 configuration, 1 device with 32 TSVs [2]
+ device_bus_width = 32
- # Default read-to-write bus around to 2 CK, @800 MHz = 2.5 ns
- tRTW = '2.5ns'
+ # HMC is a BL8 device [2]
+ burst_length = 8
- # Assume 5 CK for activate to activate for different banks
- tRRD = '6.25ns'
+ # Each device has a page (row buffer) size of 256 bytes [2]
+ device_rowbuffer_size = '256B'
- # With a 2kbyte page size, DDR3-1600 lands around 40 ns
- tXAW = '40ns'
- activation_limit = 4
+ # 1x32 configuration, so 1 device [2]
+ devices_per_rank = 1
+
+ # 4 layers so 4 ranks [2]
+ ranks_per_channel = 4
+
+ # HMC has 2 banks per layer [2]
+ # Each layer represents a rank. With 4 layers and 8 banks in total, each
+ # layer has 2 banks; thus 2 banks per rank.
+ banks_per_rank = 2
+
+ # 1250 MHz [2]
+ tCK = '0.8ns'
+
+ # 8 beats across an x32 interface translates to 4 clocks @ 1250 MHz
+ tBURST = '3.2ns'
+ # Values using DRAMSpec HMC model [1]
+ tRCD = '10.2ns'
+ tCL = '9.9ns'
+ tRP = '7.7ns'
+ tRAS = '21.6ns'
+
+ # tRRD depends on the power supply network for each vendor.
+ # We assume a tRRD of a double bank approach to be equal to 4 clock
+ # cycles (Assumption)
+ tRRD = '3.2ns'
+
+ # activation limit is set to 0 since there are only 2 banks per vault layer.
+ activation_limit = 0
+
+ # Values using DRAMSpec HMC model [1]
+ tRFC = '59ns'
+ tWR = '8ns'
+ tRTP = '4.9ns'
+
+ # Default different rank bus delay assumed to 1 CK for TSVs, @1250 MHz = 0.8
+ # ns (Assumption)
+ tCS = '0.8ns'
+
+ # Value using DRAMSpec HMC model [1]
+ tREFI = '3.9us'
+
+ # Set default controller parameters
+ page_policy = 'close'
+ write_buffer_size = 8
+ read_buffer_size = 8
+ addr_mapping = 'RoCoRaBaCh'
+ min_writes_per_switch = 8
+
+# A single DDR3-2133 x64 channel refining a selected subset of the
+# options for the DDR-1600 configuration, based on the same DDR3-1600
+# 4 Gbit datasheet (Micron MT41J512M8). Most parameters are kept
+# consistent across the two configurations.
+class DDR3_2133_x64(DDR3_1600_x64):
+ # 1066 MHz
+ tCK = '0.938ns'
+
+ # 8 beats across an x64 interface translates to 4 clocks @ 1066 MHz
+ tBURST = '3.752ns'
+
+ # DDR3-2133 14-14-14
+ tRCD = '13.09ns'
+ tCL = '13.09ns'
+ tRP = '13.09ns'
+ tRAS = '33ns'
+ tRRD = '5ns'
+ tXAW = '25ns'
+
+ # Current values from datasheet
+ IDD0 = '70mA'
+ IDD2N = '37mA'
+ IDD3N = '44mA'
+ IDD4W = '157mA'
+ IDD4R = '191mA'
+ IDD5 = '250mA'
+ VDD = '1.5V'
+
+# A single DDR4-2400 x64 channel (one command and address bus), with
+# timings based on a DDR4-2400 4 Gbit datasheet (Micron MT40A512M8)
+# in an 8x8 configuration.
+class DDR4_2400_x64(DRAMCtrl):
+ # size of device
+ device_size = '512MB'
-# A single DDR3 x64 interface (one command and address bus), with
-# default timings based on DDR3-1333 4 Gbit parts in an 8x8
-# configuration, which would amount to 4 GByte of memory. This
-# configuration is primarily for comparing with DRAMSim2, and all the
-# parameters except ranks_per_channel are based on the DRAMSim2 config
-# file DDR3_micron_32M_8B_x8_sg15.ini. Note that ranks_per_channel has
-# to be manually set, depending on size of the memory to be
-# simulated. By default DRAMSim2 has 2048MB of memory with a single
-# rank. Therefore for 4 GByte memory, set ranks_per_channel = 2
-class DDR3_1333_x64_DRAMSim2(DRAMCtrl):
# 8x8 configuration, 8 devices each with an 8-bit interface
device_bus_width = 8
- # DDR3 is a BL8 device
+ # DDR4 is a BL8 device
burst_length = 8
- # Each device has a page (row buffer) size of 1KB
- # (this depends on the memory density)
+ # Each device has a page (row buffer) size of 1 Kbyte (1K columns x8)
device_rowbuffer_size = '1kB'
# 8x8 configuration, so 8 devices
devices_per_rank = 8
- # Use two ranks
+ # Match our DDR3 configurations which is dual rank
ranks_per_channel = 2
- # DDR3 has 8 banks in all configurations
- banks_per_rank = 8
+ # DDR4 has 2 (x16) or 4 (x4 and x8) bank groups
+ # Set to 4 for x4, x8 case
+ bank_groups_per_rank = 4
- # 666 MHs
- tCK = '1.5ns'
+ # DDR4 has 16 banks (4 bank groups) in all
+ # configurations. Currently we do not capture the additional
+ # constraints incurred by the bank groups
+ banks_per_rank = 16
- tRCD = '15ns'
- tCL = '15ns'
- tRP = '15ns'
- tRAS = '36ns'
- tWR = '15ns'
- tRTP = '7.5ns'
+ # override the default buffer sizes and go for something larger to
+ # accommodate the larger bank count
+ write_buffer_size = 128
+ read_buffer_size = 64
- # 8 beats across an x64 interface translates to 4 clocks @ 666.66 MHz.
- # Note this is a BL8 DDR device.
- tBURST = '6ns'
+ # 1200 MHz
+ tCK = '0.833ns'
- tRFC = '160ns'
+ # 8 beats across an x64 interface translates to 4 clocks @ 1200 MHz
+ # tBURST is equivalent to the CAS-to-CAS delay (tCCD)
+ # With bank group architectures, tBURST represents the CAS-to-CAS
+ # delay for bursts to different bank groups (tCCD_S)
+ tBURST = '3.333ns'
- # DDR3, <=85C, half for >85C
- tREFI = '7.8us'
+ # @2400 data rate, tCCD_L is 6 CK
+ # CAS-to-CAS delay for bursts to the same bank group
+ # tBURST is equivalent to tCCD_S; no explicit parameter required
+ # for CAS-to-CAS delay for bursts to different bank groups
+ tCCD_L = '5ns';
- # Greater of 4 CK or 7.5 ns, 4 CK @ 666.66 MHz = 6 ns
- tWTR = '7.5ns'
+ # DDR4-2400 17-17-17
+ tRCD = '14.16ns'
+ tCL = '14.16ns'
+ tRP = '14.16ns'
+ tRAS = '32ns'
- # Default read-to-write bus around to 2 CK, @666.66 MHz = 3 ns
- tRTW = '3ns'
+ # RRD_S (different bank group) for 1K page is MAX(4 CK, 3.3ns)
+ tRRD = '3.3ns'
- tRRD = '6.0ns'
+ # RRD_L (same bank group) for 1K page is MAX(4 CK, 4.9ns)
+ tRRD_L = '4.9ns';
- tXAW = '30ns'
+ tXAW = '21ns'
activation_limit = 4
+ tRFC = '350ns'
+ tWR = '15ns'
+
+ # Here using the average of WTR_S and WTR_L
+ tWTR = '5ns'
+
+ # Greater of 4 CK or 7.5 ns
+ tRTP = '7.5ns'
+
+ # Default same rank rd-to-wr bus turnaround to 2 CK, @1200 MHz = 1.666 ns
+ tRTW = '1.666ns'
+
+ # Default different rank bus delay to 2 CK, @1200 MHz = 1.666 ns
+ tCS = '1.666ns'
+
+ # <=85C, half for >85C
+ tREFI = '7.8us'
+
+ # Current values from datasheet
+ IDD0 = '64mA'
+ IDD02 = '4mA'
+ IDD2N = '50mA'
+ IDD3N = '67mA'
+ IDD3N2 = '3mA'
+ IDD4W = '180mA'
+ IDD4R = '160mA'
+ IDD5 = '192mA'
+ VDD = '1.2V'
+ VDD2 = '2.5V'
# A single LPDDR2-S4 x32 interface (one command/address bus), with
-# default timings based on a LPDDR2-1066 4 Gbit part in a 1x32
-# configuration.
+# default timings based on a LPDDR2-1066 4 Gbit part (Micron MT42L128M32D1)
+# in a 1x32 configuration.
class LPDDR2_S4_1066_x32(DRAMCtrl):
+ # No DLL in LPDDR2
+ dll = False
+
+ # size of device
+ device_size = '512MB'
+
# 1x32 configuration, 1 device with a 32-bit interface
device_bus_width = 32
tRAS = '42ns'
tWR = '15ns'
- # 6 CK read to precharge delay
- tRTP = '11.256ns'
+ tRTP = '7.5ns'
# 8 beats across an x32 DDR interface translates to 4 clocks @ 533 MHz.
# Note this is a BL8 DDR device.
# Irrespective of speed grade, tWTR is 7.5 ns
tWTR = '7.5ns'
- # Default read-to-write bus around to 2 CK, @533 MHz = 3.75 ns
+ # Default same rank rd-to-wr bus turnaround to 2 CK, @533 MHz = 3.75 ns
tRTW = '3.75ns'
+ # Default different rank bus delay to 2 CK, @533 MHz = 3.75 ns
+ tCS = '3.75ns'
+
# Activate to activate irrespective of density and speed grade
tRRD = '10.0ns'
tXAW = '50ns'
activation_limit = 4
+ # Current values from datasheet
+ IDD0 = '15mA'
+ IDD02 = '70mA'
+ IDD2N = '2mA'
+ IDD2N2 = '30mA'
+ IDD3N = '2.5mA'
+ IDD3N2 = '30mA'
+ IDD4W = '10mA'
+ IDD4W2 = '190mA'
+ IDD4R = '3mA'
+ IDD4R2 = '220mA'
+ IDD5 = '40mA'
+ IDD52 = '150mA'
+ VDD = '1.8V'
+ VDD2 = '1.2V'
+
# A single WideIO x128 interface (one command and address bus), with
# default timings based on an estimated WIO-200 8 Gbit part.
class WideIO_200_x128(DRAMCtrl):
+ # No DLL for WideIO
+ dll = False
+
+ # size of device
+ device_size = '1024MB'
+
# 1x128 configuration, 1 device with a 128-bit interface
device_bus_width = 128
# Greater of 2 CK or 15 ns, 2 CK @ 200 MHz = 10 ns
tWTR = '15ns'
- # Default read-to-write bus around to 2 CK, @200 MHz = 10 ns
+ # Default same rank rd-to-wr bus turnaround to 2 CK, @200 MHz = 10 ns
tRTW = '10ns'
+ # Default different rank bus delay to 2 CK, @200 MHz = 10 ns
+ tCS = '10ns'
+
# Activate to activate irrespective of density and speed grade
tRRD = '10.0ns'
tXAW = '50ns'
activation_limit = 2
+ # The WideIO specification does not provide current information
+
# A single LPDDR3 x32 interface (one command/address bus), with
-# default timings based on a LPDDR3-1600 4 Gbit part in a 1x32
-# configuration
+# default timings based on a LPDDR3-1600 4 Gbit part (Micron
+# EDF8132A1MC) in a 1x32 configuration.
class LPDDR3_1600_x32(DRAMCtrl):
+ # No DLL for LPDDR3
+ dll = False
+
+ # size of device
+ device_size = '512MB'
+
# 1x32 configuration, 1 device with a 32-bit interface
device_bus_width = 32
# 1x32 configuration, so 1 device
devices_per_rank = 1
- # Use a single rank
+ # Technically the datasheet is a dual-rank package, but for
+ # comparison with the LPDDR2 config we stick to a single rank
ranks_per_channel = 1
# LPDDR3 has 8 banks in all configurations
# 800 MHz
tCK = '1.25ns'
- # Fixed at 15 ns
- tRCD = '15ns'
+ tRCD = '18ns'
# 12 CK read latency, 6 CK write latency @ 800 MHz, 1.25 ns cycle time
tCL = '15ns'
# Greater of 4 CK or 7.5 ns, 4 CK @ 800 MHz = 5 ns
tRTP = '7.5ns'
- # Pre-charge one bank 15 ns (all banks 18 ns)
- tRP = '15ns'
+ # Pre-charge one bank 18 ns (all banks 21 ns)
+ tRP = '18ns'
# 8 beats across a x32 DDR interface translates to 4 clocks @ 800 MHz.
# Note this is a BL8 DDR device.
# Irrespective of speed grade, tWTR is 7.5 ns
tWTR = '7.5ns'
- # Default read-to-write bus around to 2 CK, @800 MHz = 2.5 ns
+ # Default same rank rd-to-wr bus turnaround to 2 CK, @800 MHz = 2.5 ns
tRTW = '2.5ns'
+ # Default different rank bus delay to 2 CK, @800 MHz = 2.5 ns
+ tCS = '2.5ns'
+
# Activate to activate irrespective of density and speed grade
tRRD = '10.0ns'
# Irrespective of size, tFAW is 50 ns
tXAW = '50ns'
activation_limit = 4
+
+ # Current values from datasheet
+ IDD0 = '8mA'
+ IDD02 = '60mA'
+ IDD2N = '0.8mA'
+ IDD2N2 = '26mA'
+ IDD3N = '2mA'
+ IDD3N2 = '34mA'
+ IDD4W = '2mA'
+ IDD4W2 = '190mA'
+ IDD4R = '2mA'
+ IDD4R2 = '230mA'
+ IDD5 = '28mA'
+ IDD52 = '150mA'
+ VDD = '1.8V'
+ VDD2 = '1.2V'
+
+# A single GDDR5 x64 interface, with
+# default timings based on a GDDR5-4000 1 Gbit part (SK Hynix
+# H5GQ1H24AFR) in a 2x32 configuration.
+class GDDR5_4000_x64(DRAMCtrl):
+ # size of device
+ device_size = '128MB'
+
+ # 2x32 configuration, 1 device with a 32-bit interface
+ device_bus_width = 32
+
+ # GDDR5 is a BL8 device
+ burst_length = 8
+
+ # Each device has a page (row buffer) size of 2Kbits (256Bytes)
+ device_rowbuffer_size = '256B'
+
+ # 2x32 configuration, so 2 devices
+ devices_per_rank = 2
+
+ # assume single rank
+ ranks_per_channel = 1
+
+ # GDDR5 has 4 bank groups
+ bank_groups_per_rank = 4
+
+ # GDDR5 has 16 banks with 4 bank groups
+ banks_per_rank = 16
+
+ # 1000 MHz
+ tCK = '1ns'
+
+ # 8 beats across an x64 interface translates to 2 clocks @ 1000 MHz
+ # Data bus runs @2000 Mhz => DDR ( data runs at 4000 MHz )
+ # 8 beats at 4000 MHz = 2 beats at 1000 MHz
+ # tBURST is equivalent to the CAS-to-CAS delay (tCCD)
+ # With bank group architectures, tBURST represents the CAS-to-CAS
+ # delay for bursts to different bank groups (tCCD_S)
+ tBURST = '2ns'
+
+ # @1000MHz data rate, tCCD_L is 3 CK
+ # CAS-to-CAS delay for bursts to the same bank group
+ # tBURST is equivalent to tCCD_S; no explicit parameter required
+ # for CAS-to-CAS delay for bursts to different bank groups
+ tCCD_L = '3ns';
+
+ tRCD = '12ns'
+
+ # tCL is not directly found in datasheet and assumed equal tRCD
+ tCL = '12ns'
+
+ tRP = '12ns'
+ tRAS = '28ns'
+
+ # RRD_S (different bank group)
+ # RRD_S is 5.5 ns in datasheet.
+ # rounded to the next multiple of tCK
+ tRRD = '6ns'
+
+ # RRD_L (same bank group)
+ # RRD_L is 5.5 ns in datasheet.
+ # rounded to the next multiple of tCK
+ tRRD_L = '6ns'
+
+ tXAW = '23ns'
+
+ # tXAW < 4 x tRRD.
+ # Therefore, activation limit is set to 0
+ activation_limit = 0
+
+ tRFC = '65ns'
+ tWR = '12ns'
+
+ # Here using the average of WTR_S and WTR_L
+ tWTR = '5ns'
+
+ # Read-to-Precharge 2 CK
+ tRTP = '2ns'
+
+ # Assume 2 cycles
+ tRTW = '2ns'
+
+ # Default different rank bus delay to 2 CK, @1000 MHz = 2 ns
+ tCS = '2ns'
+ tREFI = '3.9us'