1
ESE 570 SEMICONDUCTOR MEMORIES
Kenneth R. Laker, University of Pennsylvania
2
Semiconductor Memories
Read-only Memory (ROM)
1. Mask programmed 2. Programmable ROM (PROM) b. Fuse ROM a. Erasable PROM (EPROM) c. Electrically Erasable PROM (EEPROM) d. Flash Memory e. Ferroelectric RAM (FRAM) Kenneth R. Laker, University of Pennsylvania
Read/Write (R/W) Memory or Random Access Memory (ROM)
Static RAM (SRAM)
Dynamic RAM (DRAM)
3
MEMORY DEVICE CHARACTERISTICS
DRAM Data volatility Data Refresh Op Cell Structure Cell Density Power Consm Read Speed Write Speed Cost per Bit Application Ex
Yes Required 1T-1C High High ≈50ns ≈40ns Low Main Memory
Kenneth R. Laker, University of Pennsylvania
SRAM
UV EPROM
Yes No No No 6T 1T Low High High/low High ≈10/70ns ≈50ns ≈5/40ns ≈10µs High Low Cashe/PDAs Game Machines
EEPROM No No 2T Low Low ≈50ns ≈5ms High ID Card
FLASH
FRAM
No No No No 1T (2 G) 1T-1C High High Low High ≈50ns ≈100ns ≈10µs-1ms ≈100ns Low Low Memory Smart Card, Card, Solid-State Digital Disk Camera
TYPICAL RANDOM ACCESS MEMORY ARRAY ORGANIZATION BIT LINES (2M) Col 2M Col 1 Col 2 A1 Row 1 A2 A3
Row 2 MEMORY CELL 2N x 2M cells
Row 2N
AN DATA LINE CONTROL CKTS
DATA CNTL N+M
CHIP I/O
COLUMN ADDRESS DECODER
Kenneth R. Laker, University of Pennsylvania
BM B1 B2 B3 COLUMN ADDRESS BITS
Memory Block
4
ROM CIRCUITS 4 X 4 BIT NOR BASED ROM ARRAY VDD VDD VDD VDD
5
BIT WORD LINES LINES R1 R2 R3 R4 C1 C2 C3 C4 1 0 0 0 0 1 0 1 0 1 0 0 0 0 1 1 0 0 1 0 1 0 0 1 0 0 0 1 0 1 1 0
R1
R2
default “0”
default “1”
R3
All word lines Ri are kept at logic “0” level, except the selected line is pulled up to “1” level.
R4
Absent Xstr => store “1” Present Xstr => store “0” C1
C2
Kenneth R. Laker, University of Pennsylvania
C3
C4
6
LAYOUT OF CONTACT - MASK PROGRAMMABLE NOR ROM
poly row (word) lines
metal column (bit) lines to load devices
metal
metal poly
R1
diffusion to GND poly
R2
To Outputs
Kenneth R. Laker, University of Pennsylvania
contact (0 bit)
NO contact (1 bit)
7
LAYOUT OF CONTACT -MASK PROGRAMMABLE 4 X 4 BIT NOR ROM
R1 R2 R3 R4
metal C1
C2
metal
metal
metal
C4
C3
poly
R1 diffusion to GND
poly
R2 metal-diff contact
poly
R3 diffusion to GND
poly
R4 Kenneth R. Laker, University of Pennsylvania
C1
C2
C3
C4
IMPLANT - MASK PROGRAMMABLE NOR ROM ARRAY Absent Xstr => store “1” Present Xstr => store “0” metal columns (bit)
poly rows (word)
threshold voltage implant to activate 1 bit VT0 > VDD permanent OFF transistor <=> contact disconnect Kenneth R. Laker, University of Pennsylvania
8
4 X 4 BIT NAND BASED ROM ARRAY VDD
C1
VDD
C2
R1 R2
VDD
C3
VDD
BIT WORD C4 LINES LINES R1 R2 R3 R4 C1 C2 C3 C4 0 1 1 1 0 1 0 1 1 0 1 1 0 0 1 1 1 1 0 1 1 0 0 1 1 1 1 0 0 1 1 0 default “1”
R3 R4
Kenneth R. Laker, University of Pennsylvania
9
default “0”
All word lines are kept at logic “1” level, except the selected line is pulled down by “0” level. Absent Xstr => store “0” Present Xstr => store “1”
10
IMPLANT - MASK PROGRAMMABLE 4 X 4 BIT NAND ROM LAYOUT Absent Xstr => store “0” (short) VDD Present Xstr => store “1” R1 R2
C1
C2
C3
C1
C4
diffusion to LOAD devices C3 C2
C4 poly
R1
R3 R4
poly R2 poly R3
poly R4 diffusion to GND
Kenneth R. Laker, University of Pennsylvania
threshold voltage implant to make VT0 negative to store “0”
DESIGN OF ROW AND COLUMN DECODERS ROW ADDRESS DECODER FOR 2 ADDRESS BITS AND 4 WORD LINES EXAMPLE RA1
ROW DECODER
RA2 RA1 RA2 0 0 0 1 1 0 1 1 2 address bits plus complements Kenneth R. Laker, University of Pennsylvania
R1 R2 R3 R4
R1 R2 R3 R4 1 0 0 0 0 1 0 0 0 0 1 0 0 0 0 1 Select 1 of 4 rows or word lines
11
NOR BASED IMPLEMENTATION FOR A ROW DECODER WITH 2 ADDRESS BITS VDD AND 4 WORD LINES R1 off
VDD
off
(2N) nMOS + (N )pMOS = (2N + N) Xstrs VDD PLUS 2N INVERTERS
R2 off
on
R3 on
off
R4
0 RA2 1 0 RA2
on
0 RA1 1 0 RA1
Kenneth R. Laker, University of Pennsylvania
12
on
VDD
For N = 2 RA1 RA2 R1 R2 R3 R4 0 0 1 0 0 0 0 1 0 1 0 0 1 0 0 0 1 0 1 1 0 0 0 1
13
REALIZATION OF ROW DECODER AND ROM ARRAY AS TWO ADJACENT NOR PLANES 2N ROWS (WORD LINES) R1 R2 NOR ROW NOR ROM DECODER ARRAY R2N
RA1 RA2
RAN
N ROW ADDRESS BITS
Kenneth R. Laker, University of Pennsylvania
C2M C1 C2 2M COLUMNS (BIT LINES) For N = 2 RA1 RA2 0 0 0 1 1 0 1 1
R1 R2 R3 R4 1 0 0 0 0 1 0 0 0 0 1 0 0 0 0 1
REALIZATION OF ROW DECODER AND ROM ARRAY AS TWO ADJACENT NAND PLANES 2N ROWS (WORD LINES) R1 R2 NAND NAND ROW ROM DECODER ARRAY N R2
RA1 RA2 RAN N ROW ADDRESS BITS RA1 RA2 0 0 0 1 1 0 1 1 Kenneth R. Laker, University of Pennsylvania
C2M C1 C2 2M COLUMNS (BIT LINES)
R1 R2 R3 R4 0 1 1 1 1 0 1 1 1 1 0 1 1 1 1 0
ROW DECODER TRUTH TABLE FOR A 4X4 NAND ROM ARRAY
14
COLUMN DECODER SCHEME USING NOR ADDRESS DECODER AND nMOS PASS TRANSISTORS
15
ROM ARRAY
CA1 CA2
NOR COLUMN ADDRESS DECODER
C1 C2 C3
2M pass transistors
(2M+M)Transistors
C2M
CAM M column address bits
SAME AS ROW DECODER
Kenneth R. Laker, University of Pennsylvania
DATA OUTPUT serial (as shown)
COLUMN DECODER FOR 8 BIT LINES IMPLEMENTED AS BINARY16 TREE C2 C3 C4 C5 C6 C7 C8 C1
COLUMN ADDRESS BITS
DATA OUTPUT serial or parallel ADVG: Decoding is realized by tree structure; much reduced Xstr count. DISADVG: Large number of series connected nMOS pass Xstrs. Fix with buffers or use of a combined tree and pass-transistor design. Kenneth R. Laker, University of Pennsylvania
EXAMPLE 10.1: Consider the design of a 32 - kbit implant-mask programmable NOR ROM (215 = 32,768 individual memory cells) and the design issues that 4µ/1.5µ relate to ACCESS TIME. NOTE: row address bits + col address bits = 15 R1 ASSUME: row access bits = 7 i.e. 27 = 128 rows, col access bits = 8, i.e. 28 = 256 cols. R2 Estimate: row access time, column access R127 time and total access time. row R128 4µ 2µ 2µ 4µ 6µ
17
2µ/1.5µ
C1
C2
C255
C256
column n+ diffusion Process Params high threshold imp µnCox = 20µA/V2 “unit” memory cell (W = 2µm, L = 1.5µm) Cox = 3.47fF/µm2 Rsheet-poly = 20Ω/sq Cgdn+Cdbn = 0.0118pF n+ diffusion
1. Calculate Rrow, Crow for Unit Memory Cell: Crow = Cox W L = 3.47 fF (2 x 1.5) = 10.4 fF per cell Rrow = Rsheet-poly (Lp/Wp) = 20 Ω (3) = 60 Ω per cell Kenneth R. Laker, University of Pennsylvania
18
row R128
R, C for Unit Memory Cell, i.e. per Bit Crow = Cox W L = 3.47 fF (2 x 1.5) = 10.4 fF per cell Rrow = Rsheet-poly (L/W) = 20 Ω (3) = 60 Ω per cell
C1
C2
C255
C256
RC TRANSMISSION LINE MODEL FOR POLY WORD LINE (ROW) WITH UP TO 256 TRANSISTORS (# OF COLUMNS) R1 R2 R255 R256 VG256 R = R Vin i row Ci = Crow C1 C2 C256 C255 2. Calculate row access time trow -> delay associated with selecting and activating 1 of 128 word lines in ROM array V Emperical Delay Formulas VOH Vin V256 V50% t trow Kenneth R. Laker, University of Pennsylvania
= 15.5 ns (at VG256)
3. Calculate column access time tcolumn -> worst case delay τPHL associated with discharging the precharged bit line when a row is activated.
19
R1 R2
Up to 128-input NOR gate VDD rep of column in the ROM (4/1.5) array.
R1
R2
R3
R127
column output CJ
R128
(2/1.5)
Ccolumn
Ccolumn = 128 x (Cgd,n + Cdbn) = 1.5 pF where it is estimated that Cgdn + Cdbn = 0.0118 pF per word line
Kenneth R. Laker, University of Pennsylvania
R128 C1
Ccolumn = 128 x (Cgd,n + Cdbn) = 1.5 pF
20
where it is estimated that Cgdn + Cdbn = 0.0118 pF per word line Since only one word-line (row) is activated at a time, tcolumn = τPHL (falling output voltage) is determined by the inverter circuit. Output capacitance Ccolumn is assumed pre-charged to near VDD before each row access operation. VDD (4/1.5)
R1
column output CJ
(2/1.5)
Ccolumn = 1.5 pF
tcolumn = τPHL = 18 ns CC column
τ PHL
2 VT 0 n 4(VOH − VT 0 n ) = + ln − 1 = 18 ns k n (VOH − VT 0 n ) VOH − VT 0 n VOH + VOL load
taccess = trow + tcolumn = 15.5 ns + 18 ns = 33.5 ns Kenneth R. Laker, University of Pennsylvania
STATIC READ-WRITE MEMORY (SRAM) CIRCUITS bit line bit line (column)
21
(column)
1 - BIT SRAM CELL Complementary Column arrangement achieves more reliable operation. bit line (column)
VDD load
load
bit line (column)
word line word line (row) (row) BASIC REQUIREMENTS THAT DICTATE DESIGN: 1. DATA-WRITE OP -> MODIFY STORED DATA IN SRAM CELL 2. DATA-READ OP -> NOT MODIFY STORED DATA IN SRAM CELL Kenneth R. Laker, University of Pennsylvania
22
RESISTIVE-LOAD SRAM CELL VDD bit line
bit line R
word line
pass transistors to activated by a row select (RS) signal to enable read/write ops
Kenneth R. Laker, University of Pennsylvania
R
word line
basic cross-coupled 2-inverter latch with 2 stable op points for storing one-bit
SRAM cell is accessed via two bit (column) lines C and its complement for more reliable and much faster operation
23
VDD bit line
bit line Full CMOS SRAM Cell
word line
Kenneth R. Laker, University of Pennsylvania
word line
VDD MP1
SRAM OPERATION PRINCIPLES pull-up transistors (one per column) VDD
bit line R VC
M3
CC
V1 M1
R V2 M4 M2
24
VDD MP2 bit line VNOT-C CNOT-C
RS
word line 1. WHEN THE WORD LINE NOT SELECTED, i.e. RS = 0: M3 & M4 are OFF a. One data-bit is HELD, i.e. latch preserves one of its two stable states. b. If RS = 0 for ALL rows, the bit lines capacitances CC and CNOT-C are charged-up to near VDD by pull-up of MP1 and MP2 (both in SAT). VNOT-C = EX: VC = VNOT-C = 3.5 V for VDD = 5 V, VT0 = 1 V, |2φF| = 0.6 V, γ = 0.4 V1/2. Kenneth R. Laker, University of Pennsylvania
VDD MP1
MP2 bit line
bit line R VC
M3
CC
V1 M1
25
VDD
pull-up transistors (one per column) VDD R V2 M4 M2
VNOT-C CNOT-C
RS 2. WHEN THE WORD LINE IS SELECTED, i.e. RS = 1: M3 & M4 are ON Four basic ops can be performed in this SRAM Mode: a. WRITE “1” OP (@ t = 0− V1n-1 = VOL, V2n-1 = VOH) RS -> 1 at t = 0: VNOT-C -> VOL by the DATA-WRITE CIRCUITRY. Hence V2n -> VOL, then M1 turns OFF => V1n -> VOH and M2 turns ON pulling V2n -> VOL. b. READ “1” OP (@ t = 0− V1n-1 = VOH, V2n-1 = VOL) RS -> 1 at t = 0: VC retains precharge level, while VNOT-C -> VOL by ON M2. DATA-READ CIRCUITRY detects small voltage difference VC - VNOT-C > 0, and amplifies it as a “1” data output. Kenneth R. Laker, University of Pennsylvania
VDD MP1
MP2 bit line
bit line R VC CC
M3
V1 M1
26
VDD
pull-up transistors (one per column) VDD R V2 M4 M2
VNOT-C CNOT-C
RS 2. WHEN THE WORD LINE IS SELECTED, i.e. RS = 1: M3 & M4 are ON Four basic ops can be performed in this SRAM Mode: c. WRITE “0” OP (@ t = 0− V1n-1 = VOH, V2n-1 = VOL) RS -> 1 at t = 0: VC -> VOL by the DATA-WRITE CIRCUITRY. Since V1 -> VOL, M2 turns OFF => V2 -> VOH and M1 turns ON pulling V1n -> VOL. d. READ “0” OP (@ t = 0− V1n-1 = VOL, V2n-1 = VOH) RS -> 1 at t = 0: VNOT-C retains precharge level, while VC -> VOL by ON M1. DATA-READ CIRCUITRY detects small voltage difference VC - VNOT-C < 0, and amplifies it as a “0” data output. Kenneth R. Laker, University of Pennsylvania
VDD
pull-up transistors (one per column) VDD
VDD MP1
MP2 bit line
bit line R VC
M3
CC
R V2 M4
V1
27
VNOT-C
M2
M1
CNOT-C
RS write “1” hold
hold
VC ≈ 3.5 V
Precharge
RS
write “0”
read “1” V1 = VOH V2 = VOL
hold
hold Precharge
≈0V
≈ 3.5 V
Precharge
VNOT-C
≈0V Kenneth R. Laker, University of Pennsylvania
Precharge
≈ 3.0 V
read “0”
Precharge Precharge
V1 = VOL V2 = VOH
hold
≈ 3.0 V
≈ 3.5 V
STATIC OR “STANDBY” POWER CONSUMPTION VDD pull-up transistors VDD (one per column) MP2 MP1 VDD bit line bit line R R VC M3 V1 V2 M4 VNOT-C CC
M1
M2
28
CNOT-C
RS ASSUME: 1 bit is stored in the cell => M1 OFF, M2 ON => V1 = VOH, V2 = VOL. i.e. ONE LOAD RESISTOR IS ALWAYS CONDUCTING NON-ZERO CURRENT. with R = 100 MΩ (undoped poly), Pstandby < 0.25 µW per cell for VDD = 5V Kenneth R. Laker, University of Pennsylvania
VDD MP1 bit line VC CC
29
CIRCUIT FOR CMOS SRAM CELL psedo-nMOS pull-up VDD transistors (one per column) MP2 VDD bit line M5
M6
M1
M2
M3
RS
M4
VNOT-C CNOT-C
word line -> VERY LOW STANDBY POWER CONSUMPTION -> LARGER NOISE MARGINS THAN R-LOAD SRAMS -> OPERATE AT LOWER SUPPLY VOLTAGES THAN R-LOAD SRAMS -> LARGER DIE AREA -> CMOS MORE COMPLEX FAB PROCESS Kenneth R. Laker, University of Pennsylvania
CMOS SRAM DESIGN STRATEGY BASIC REQUIREMENTS THAT DICTATE (W/L) RATIOS: 1. DATA-WRITE OP -> MODIFY STORED DATA IN SRAM CELL 2. DATA-READ OP -> NOT MODIFY STORED DATA IN SRAM CELL 1. CONSIDER DATA-READ OP with “0” STORED IN CELL: VDD VDD MP2 bit line
VDD
MP1 bit line
M5 M3 0V
CC
M1
V1
V2
M6 VDD M4 M2
RS RS = VDD a. @ t = 0−: M3, M4 OFF; M2, M5 OFF & M1, M6 LIN b. @ t = 0: M3 SAT, M4 LIN; M2, M5 OFF & M1, M6 LIN slow discharge of large Cc and V1 increases REQUIRE V1 < VT02 => LIMITS M3 W/L wrt M1 W/L Kenneth R. Laker, University of Pennsylvania
CNOT-C
30
VDD
VDD
MP2 bit line
VDD
MP1 bit line
M5 M3 0V
CC
V2
V1
31
M6 VDD M4 CNOT-C
M2
M1 RS
RS = VDD DESIGN CONSTRAINT: i.e. KEEP M2 OFF M3 SAT, M1 LIN => kn 3 2 k n1 2( VDD − VT 0 n ) V1 − V12 ( VDD − V1 − VT 0 n ) = 2 2
(
Kenneth R. Laker, University of Pennsylvania
)
SYMMETRY: kn 4 SAME for kn 2 (M1 & M6 OFF for Read ‘1”)
2. CONSIDER DATA-WRITE “0” OP with “1” STORED IN CELL: VDD VDD 0V bit line bit line M6 M5 0V M4 V 0V V2 M3 DD V1 CNOT-C CC M2 M1 RS
RS = VDD at t =0
VC IS SET “0” BY DATA-WRITE CIRCUIT a. @ t = 0−: M3, M4 OFF; M2, M5 LIN & M1, M6 OFF (“1” stored) b. @ t = 0: M3 SAT, M4 SAT; M2, M5 LIN & M1, M6 OFF WRITE “0” => V1: VDD -> 0 (< VT0n) AND V2: 0 -> VDD (M2 -> OFF) i.e. KEEP M2 OFF
DESIGN CONSTRAINT:
WHEN V1 = VT0n: M3 LIN & M5 SAT => kp 5 2 kn 3 0 − VDD − VT 0 p = 2( VDD − VT 0 n ) VT 0 n − VT 0n 2 2 2
(
Kenneth R. Laker, University of Pennsylvania
)
(
)
32
33
V1 < VT0n, i.e. M2 (M1) forced OFF kp 5 kn 3
=
kp 6 kn 4
<
2(VDD −1.5VT 0 n ) VT 0 n
(VDD + VT 0p )
=>
2
BY SYMMETRYto WRITE “1” when “0” is stored in cell. W W L 5 L 6 µ n 2( VDD − 1.5VT 0 n ) VT 0 n = < 2 W W µp VDD + VT 0 p L 3 L 4
(
Kenneth R. Laker, University of Pennsylvania
)
VDDSRAM WRITE CIRCUIT VDD
MP1 SHARED BY SEVERAL COLUMNS
bit line
VDD
VC
1 - bit SRAM Cell
VNOT-C
34
MP2 bit line
RS WB
word line M2
DATA
M1 from COLUMN DECODER
M3
Write OP: force VC or VNOT-C to a logic low level when W = 0. NOT-W DATA NOT-WB WB 0 1 0 1 0 0 1 0 1 X 0 0 Kenneth R. Laker, University of Pennsylvania
OPERATION (M3 ON) M1 OFF, M2 ON ->VNOT-C LOW M1 ON, M2 OFF ->VC LOW M1 OFF, M2 OFF ->VC = VNOT-C HIGH
SRAM READ CIRCUIT
35
VDD
SOURCE COUPLED DIFFERENTIAL VC AMPLIFIER
R
Vo1 Vo2 M1 Vx M2
R
VNOT-C
ISS
Asense =
Kenneth R. Laker, University of Pennsylvania
Increase R -> Use active loads Use cascode
36
FAST SENSE AMPLIFIERS VDD
VC
M4
pMOS current M5 nirror
M1
M2 VNOT-C
CC
VDD Vo
VoN
CNOT-C CK bit line
M3 bit line
Asense = -gm2 (ro2||ro5)
Kenneth R. Laker, University of Pennsylvania
37 DYNAMIC READ-WRITE MEMORY (DRAM) CIRCUITS SRAM -> 4 - 6 TRANSISTORS PER BIT 3 - 5 LINES CONNECTING EACH CELL DRAM -> DATA BIT IS STORED AS CHARGE ON CAPACITOR REDUCED DIE AREA VS. SRAM REQUIRES PERIODIC REFRESH bit line bit line 2 parasitic storage 4 - Transistor capacitors DRAM Cell (earliest cell, C C word word based on 6 Xstr line line SRAM cell) CC and CNOT-C >> C RS (read select) bit line (write) 1 parasitic storage capacitor bit line (read) 3 - Transistor DRAM Cell C (first widely WS (write select) used cell) Kenneth R. Laker, University of Pennsylvania
38
word line (read/write select)
bit line (data read/write)
explicit C storage capacitor
1 - Transistor DRAM Cell (most widely used cell)
-> INDUSTRY STANDARD FOR HIGH DENSITY DRAM ARRAYS -> SMALLEST COMPONET COUNT & SILICON AREA PER BIT -> SEPARATE OR “EXPLICIT” CAPACITOR (DUAL POLY) PER CELL
Kenneth R. Laker, University of Pennsylvania
OPERATION OF 3 - TRANSISTOR DRAM CELL
39
VDD MP1
Precharge devices
1 - Bit DRAM Cell
PC
read select (RS)
CC
MP2
M1
M2
M3 CNOT-C
C write select (WS) DATA
bit line Data_in
bit line Data_out
CC, CNOT-C >> C (> 10 C) Uses two-phase non-overlapping clock scheme where φ1 = PC = precharge and φ2 = RS = read or WS = write. Kenneth R. Laker, University of Pennsylvania
VDD MP1
Precharge devices
MP2 1 - Bit DRAM Cell
PC
read select (RS)
CC
40
M3
M1
M2
CNOT-C
C write select (WS) DATA
bit line Data_in
bit line Data_out
PC WS DATA Data_in Stored Data RS Data_out Kenneth R. Laker, University of Pennsylvania
read write read write PC “1” PC “1” PC “0” PC “0” 3 5 4 7 8 1 2 6
41
VDD MP1
Precharge devices
MP2 PRECHARGE CYCLE
PC CNOT-C
CC
RS VDD - VTMP1 M1
M2
CC
M3 C3
C WS DATA
bit line Data_in
Kenneth R. Laker, University of Pennsylvania
bit line Data_out
WRITE “1” OP DATA = 0 WS = 1; RS = 0 CC, C SHARE CHARGE DUE TO M1 ON
42
RS
VDD - VTMP1 M1
C2
M2
M3
C WS DATA
bit line Data_in
bit line Data_out
READ “1” OP CNOT-C DATA = 0 WS = 0; RS = 1 CNOT-C DISCHARGES THROUGH M2 AND M3
RS VDD - VTMP1 CC
M1
M2
M3 C3
C WS DATA
bit line Data_in
Kenneth R. Laker, University of Pennsylvania
bit line Data_out
WRITE “0” OP DATA = 1 WS = 1; RS = 0 CC, C DISCHARGED TO 0 THROUGH M1 AND DATA-IN nMOS
43
RS
C2
VDD - VTMP1 M1
0V
M2
M3
C WS DATA
bit line Data_in
Kenneth R. Laker, University of Pennsylvania
bit line Data_out
READ “0” OP CNOT-C DATA = 1 WS = 0; RS = 1 CNOT-C DOES NOT DISCHARGE DUE TO M2 OFF
1 - TRANSISTOR DRAM Cell R (R/W select)
Bit-line M1
CC
C
1-Bit DRAM Cell
D (data in/out) CC >> C Uses two-phase non-overlapping clock scheme where φ1 = bit-line is precharged and φ2 = R = read/write. WRITE “1” OP: D = 1, R = 1 (M1 ON) => C CHARGES TO “1” WRITE “0” OP: D = 0, R = 1 (M1 ON) => C DISCHARGES TO “0” READ OP: DESTROYS STORED CHARGE ON C => REFRESH IS NEEDED AFTER EVERY DATA READ OP
Kenneth R. Laker, University of Pennsylvania
44
45
SRAM vs. DRAM + trans
+ o +
+ o o
Kenneth R. Laker, University of Pennsylvania
trans