1. Theory

Definition: An electrical fuse (eFuse) is a narrow silicided polysilicon or metal link that is programmed once from low to high resistance by a current pulse; an antifuse is a thin dielectric, usually a gate oxide, that is programmed once from an open to a conducting state by breaking it down (Kothandaraman et al.; Sidense).
One-time programmable storage: Both store bits permanently after manufacture, for redundancy repair, trimming, configuration and keys, without the extra masks of floating-gate memory (Kothandaraman et al.; Sidense).
eFuse states: A silicided polysilicon fuse rises from about 100 $Ω$ to well above 1 $kΩ$ , a few megohms in the 0.12 $μ m$ device of Kothandaraman et al., through electromigration of the silicide (Kothandaraman et al.; US 8,421,186).
Antifuse states: An unprogrammed oxide cell is a capacitor, and a programmed one conducts through a breakdown spot, with a read current several orders of magnitude higher (Sidense; Han et al.).
Programming cost: eFuses need milliamperes for microseconds to hundreds of microseconds, while antifuses need about 5 to 6 $V$ across a 2 $nm$ oxide but little current (Kothandaraman et al.; Sidense).

2. Working Principle

Electromigration programming: The current pulse heats the link to about 850 $^{\circ} C$ , and the electron wind drives the cobalt silicide from the cathode to the anode, leaving a high-resistance polysilicon gap (Kothandaraman et al.).
Controlled versus rupture programming: A controlled pulse moves the silicide without melting the link, while too strong a pulse ruptures it in microseconds; metal fuses form a void at the hot centre in the same way (Kothandaraman et al.; US 8,421,186).
Oxide breakdown programming: A high field generates defects in the oxide until a percolation path forms, and the current then forms a conductive filament whose resistance depends on the power allowed during breakdown (Sidense; US 7,009,891; background knowledge for percolation).
Read: An eFuse is read by comparing its resistance with a reference at low current, and an antifuse by applying a low voltage and sensing the current through the breakdown spot, often through a diode or select transistor (Kothandaraman et al.; Han et al.).

3. Applications

Redundancy repair: Replacing faulty rows and columns in embedded DRAM and SRAM after test, as in the 16 Mb eDRAM of Kothandaraman et al. (Kothandaraman et al.).
Trimming and calibration: Setting analog references, oscillator frequencies and sensor offsets in mixed-signal and power ICs, for example in BCD processes (Jin et al.; background knowledge).
Code storage and security: Antifuse OTP for boot code, configuration and keys, since its contents are hard to read by de-processing (Sidense).
Chip identification and configuration: Chip IDs, feature enables and speed bins (background knowledge).

Electrical Fuses

Name	Formula	Related Formulae	Variables/Constants
Resistance change and sensing of an eFuse	$\frac{R _{p os t}}{R _{p r e}} ≫ 1, V_{se n se} = I_{se n se} R_{f u se} compared with I_{se n se} R_{r e f}$	$R_{p r e} < R_{r e f} < R_{p os t}$	$R_{p r e}, R_{p os t} : f u se r es i s t an ce b e f or e an d a f t er p r o g r ammin g$ $V_{se n se} : se n se d v o l t a g e$ $I_{se n se} : se n se c u r r e n t$ $R_{f u se} : f u se r es i s t an ce$ $R_{r e f} : r e f er e n ce r es i s t an ce (se n se co m p a r i so n, ba c k g r o u n d k n o w l e d g e); K o t han d a r aman e t a l . (0.12 μ m C M O S, co ba l t - s i l i c i d e d p o l y s i l i co n l ink o f o pt im u m l e n g t h 1.2 μ m in t h e s t an d a r d g a t e s t a c k w i t h n o a dd e d s t e p s) : ab o u t 120 Ω b e f or e an d a f e w MΩ a f t er p r o g r ammin g (t h e p r in t e d v a l u e i s g a r b l e d, m y r e a d in g o f t h e f i g u r e c a pt i o n), l a r g e e n o ug h f or s im pl e se n se c i r c u i t s, w i t h t h e l o w se n se c u r r e n t p r e v e n t in g r e p r o g r ammin g; u se d f or r e d u n d an cy in a 16 M b e mb e dd e d D R A M; p r o g r amm e d an d u n p r o g r amm e d f u ses p a sse d r e l iabi l i t y t es t s; U S 8, 421, 186 : u n p r o g r amm e d e F u ses a r e t y p i c a l l y b e l o w 100 Ω an d p r o g r amm e d o n es ab o v e 1 kΩ$ $(co m p u t e d : 120 Ω t o an a ss u m e d 5 MΩ i s a r a t i o o f ab o u t 4 \times 1 0^{4})$
Programming pulse and energy	$E_{p r o g} \leq V_{F S} I_{p e ak} t_{p}, P = I^{2} R$	$Δ T \approx P R_{t h}$	$E_{p r o g} : p r o g r ammin g e n er g y$ $V_{F S} : f u se s u ppl y v o l t a g e$ $I_{p e ak} : p e ak c u r r e n t$ $t_{p} : p u l se l e n g t h$ $P : d i ss i p a t e d p o w er$ $I : c u r r e n t$ $R : l ink r es i s t an ce$ $Δ T : t e m p er a t u r e r i se$ $R_{t h} : t h er ma l r es i s t an ce o f t h e l ink (ba c k g r o u n d k n o w l e d g e f or t h e f or m s); K o t han d a r aman e t a l . : a 1.5 V, 200 μ s g a t e p u l se o n t h e p r o g r ammin g N F E T w i t h a 3.3 V f u se s u ppl y g i v es ab o u t 10 mA p e ak, l imi t e d b y t h e N F E T; t h e l ink r es i s t an ce r i ses t o ab o u t 360 Ω d u r in g t h e p u l se, t h e l ink r e a c h es ab o u t 850^{\circ} C, w h er e t h e p o l y s i l i co n b eco m es co n d u c t i v e, t h e c u r r e n t i s s t ab l e f or ab o u t 150 μ s an d t h e v o l t a g e t u r n s a r o u n d n e a r 220 μ s; ab o v e a 1.5 V g a t e d r i v e t h e p u l se l a s t s o n l y 1 t o 5 μ s an d t h e f u se r u pt u r es$ $(co m p u t e d : a t m os t 3.3 V x 10 mA x 200 μ s = 6.6 μ J; 10 mA in 360 Ω i s 36 mW, so 825 K o f h e a t in g im pl i es R_{t h} o f ab o u t 25 K/mW)$
Electromigration in the fuse link	$v_{d} = \frac{D}{k _{B} T} Z^{*} e ρ j, D = D_{0} exp (- \frac{E _{a}}{k _{B} T})$	$j L > (j L)_{c} (Blech condition for migration), L > 3 L_{B}$	$v_{d} : d r i f t v e l oc i t y o f t h e mi g r a t in g a t o m s$ $D : d i f f u s i v i t y$ $D_{0} : p r e f a c t or$ $E_{a} : a c t i v a t i o n e n er g y$ $k_{B} : B o l t z mann co n s t an t$ $T : t e m p er a t u r e$ $Z^{*} : e f f ec t i v e c ha r g e n u mb er$ $e : e l e m e n t a r y c ha r g e$ $ρ : r es i s t i v i t y$ $j : c u r r e n t d e n s i t y$ $L : l in e l e n g t h, (j L)_{c} : cr i t i c a l B l ec h p r o d u c t$ $L_{B} : B l ec h l e n g t h (ba c k g r o u n d k n o w l e d g e; e l ec t r o mi g r a t i o n o n t h e M e t a l an d I n t er co nn ec t p a g e); K o t han d a r aman e t a l . : co ba l t i s d e pl e t e d n e a r t h e c a t h o d e an d p i l es u p a t t h e an o d e, t h e p a tt er n r e v er ses w i t h c u r r e n t d i r ec t i o n, an d a l a r g er c a t h o d e r ai ses t h e t e m p er a t u r e g r a d i e n t, w hi c h d r i v es mi g r a t i o n m or e t han mi cr os t r u c t u r e d oes; t u n g s t e n s i l i c i d e b e ha v es t h e s am e; U S 8, 421, 186 : a B E O L m e t a l e F u se o f minim u m w i d t h (ab o u t 45 nm) an d a s p ec t r a t i o ab o v e 5, l o n g er t han t h r ee B l ec h l e n g t h s, p r o g r amm e d b y e l ec t r o mi g r a t i o n r a t h er t han b l o w o u t, f or m s a ce n t r a l v o i d w i t h a m or e r e p r o d u c ib l e, hi g h er f ina l r es i s t an ce an d n ee d s m u c h l ess a r e a t han a s i l i c i d e f u se$ $(n o co m p u t e d v a l u e)$

Gate-Oxide Antifuses

Name	Formula	Related Formulae	Variables/Constants
Oxide field and breakdown current in an antifuse	$E_{o x} = \frac{V _{p r o g}}{t _{o x}}$	$J_{B D} = \frac{I _{B D}}{A _{s p o t}}$	$E_{o x} : o x i d e f i e l d$ $V_{p r o g} : p r o g r ammin g v o l t a g e a cr oss t h e o x i d e$ $t_{o x} : o x i d e t hi c k n ess$ $J_{B D} : c u r r e n t d e n s i t y in t h e b r e ak d o w n s p o t$ $I_{B D} : b r e ak d o w n c u r r e n t$ $A_{s p o t} : s p o t a r e a; S i d e n se : ab o u t 6 V f or a 2 nm o x i d e, or 30 MV/cm, a b r e ak d o w n c u r r e n t o f ab o u t 100 μ A p er 100 n m^{2} f or min g t h e co n d u c t i v e c hann e l, b r e ak d o w n t y p i c a l l y in 100 μ s or l ess, r e a d a ccess a s l o w a s 10 ns, an d a 1 T s pl i t - c hann e l ce l l t ha t co n f in es b r e ak d o w n t o t h e c hann e l an d i s n o t m u c h l a r g er t han a ma s k - R O M ce l l; U S 7, 009, 891 : an an t i f u se c a p a c i t or o f ab o u t 20 \overset{˚}{A} b r e ak in g d o w n a t ab o u t 4.6 V an d p r o g r amm e d w i t h ab o u t 6 V e f f ec t i v e (2.5 V o n t h e g a t e, - 3.5 V o n t h e w e l l), w i t h p o w er l imi t e d t o a v o i d v o i d in g t h e o x i d e; H an e t a l . : b r e ak d o w n a t ab o u t 4.4 V an d p r o g r ammin g a t 5.0 V o n t h e w or d l in e$ $(co m p u t e d : 6 V o n 2 nm i s 30 MV/cm; 4.6 V o n 2 nm i s 23 MV/cm; 100 μ A in 100 n m^{2} i s 1 \times 1 0^{8} A/c m^{2})$
Time to breakdown and program disturb	$t_{B D} \propto V^{- n} (power law), \frac{t _{B D} ( V _{1} )}{t _{B D} ( V _{2} )} = (\frac{V _{2}}{V _{1}})^{n}$	$t_{B D} \propto exp (- γ V) (exponential law)$	$t_{B D} : t im e t o b r e ak d o w n$ $V : o x i d e v o l t a g e$ $n : v o l t a g e a cce l er a t i o n e x p o n e n t$ $γ : e x p o n e n t ia l a cce l er a t i o n f a c t or$ $V_{1}, V_{2} : tw o v o l t a g es (ba c k g r o u n d k n o w l e d g e; T D D B o n t h e E l ec t r i c a l S t r ess p a g e); H an e t a l . : u n se l ec t e d ce l l s see 2.5 V in 60 μ s p u l ses d u r in g p r o g r ammin g, r ai s in g t h e i r u n p r o g r amm e d c u r r e n t b y u p t o t e n t im es t h r o ug h s t r ess - in d u ce d l e ak a g e, an d t h e o p er a t in g co n d i t i o n g u a r an t ees 1 0^{3} w r i t e cy c l es$ $(co m p u t e d : w i t h an a ss u m e d n = 40, a ce l l t ha t b r e ak s d o w n in 100 μ s a t 5 V w o u l d l a s t ab o u t 1.1 \times 1 0^{8} s, 3.5 y e a r s, a t 2.5 V)$
Programmed resistance and read current of an antifuse	$R_{p r o g} = \frac{V _{r e a d}}{I _{r e a d}}, \frac{I _{o n}}{I _{o f f}}$	$N_{ma x} ≲ \frac{I _{o n}}{I _{o f f, r e v}} (cells sharing a line)$	$R_{p r o g} : r es i s t an ce a f t er p r o g r ammin g$ $V_{r e a d} : r e a d v o l t a g e$ $I_{r e a d} : r e a d c u r r e n t$ $I_{o n}, I_{o f f} : r e a d c u r r e n t s o f p r o g r amm e d an d u n p r o g r amm e d ce l l s$ $N_{ma x} : n u mb er o f ce l l s w h ose r e v er se l e ak a g e c an s ha r e a se n se l in e$ $I_{o f f, r e v} : r e v er se c u r r e n t o f u n se l ec t e d ce l l s; U S 7, 009, 891 : p r o g r amm e d an t i f u ses o f h u n d r e d s o f Ω or l ess, ab o u t f o u r or d er s o f ma g ni t u d e b e l o w t h e u n p r o g r amm e d s t a t e, a g ain s t a f e w kΩ t o 10 t o 100 kΩ f or co n v e n t i o na l an t i f u ses, r e a d a t 0.1 V; J in e t a l . (1 k bi t O T P in 0.18 μ m B C D, 567.9 b y 205.1 μ m) : a p os t - p r o g r am r es i s t an ce o f se v er a l t e n s o f kΩ an d s t ab l e se n s in g a t or b e l o w t ha t; H an e t a l . : ab o u t 1 μ A p r o g r amm e d an d 1 pA u n p r o g r amm e d a t 1 V r e a d, w i t h a f or w a r d - t o - r e v er se r a t i o o f ab o u t 1 0^{3} t ha t se t s t h e ma x im u m a r r a y s i z e, a n e g l i g ib l e r e a d - w in d o w c han g e a t 85^{\circ} C an d a w in d o w ab o v e f o u r or d er s o f ma g ni t u d e a f t er a 200^{\circ} C r e t e n t i o n bak e$ $(co m p u t e d : 1 μ A a t 1 V i s an e f f ec t i v e 1 MΩ in c l u d in g t h e j u n c t i o n; an o n - o f f r a t i o o f 1 \times 1 0^{6})$

Important Questions

Ques: How does an eFuse change from about 100 ohms to megohms?

Kothandaraman et al. programmed a cobalt-silicided polysilicon link of 1.2 $μ m$ length in 0.12 $μ m$ CMOS with a 200 $μ s$ pulse from a 3.3 $V$ supply, limited to about 10 $mA$ by the programming NFET. The link heats to about 850 $^{\circ} C$ and its resistance rises to about 360 $Ω$ during the pulse, while the electron wind drives cobalt from the cathode to the anode. When the silicide has left the hot region, the remaining polysilicon gap raises the resistance from about 120 $Ω$ to a few $MΩ$ . The energy is at most 6.6 $μ J$ , and 36 $mW$ of heating for an 825 $K$ rise implies a thermal resistance of about 25 $K/mW$ (computed). Above 1.5 $V$ of gate drive the link ruptures within 1 to 5 $μ s$ instead. Reference: C. Kothandaraman, S. K. Iyer and S. S. Iyer, ‘Electrically Programmable Fuse (eFUSE) Using Electromigration in Silicides’, IEEE Electron Device Letters 23(9), 523-525 (2002) (final resistance garbled in the text) (fetch-tool summary).

Ques: Is a programmed eFuse above 1 kilohm or several megohms?

Both values appear, and they describe different stages of the technology. US 8,421,186 gives a general rule of below 100 $Ω$ unprogrammed and above 1 $kΩ$ programmed, a minimum that sense circuits must resolve over all fuses and conditions. Kothandaraman et al. report about 120 $Ω$ before and a final value that the figure caption places at a few megohms; the printed number is garbled in the text (my reading). A single well-programmed silicide fuse therefore reaches megohms, while the 1 $kΩ$ value is the guaranteed lower limit for the whole population, including partially programmed fuses (my reading). The patent argues that its electromigration-programmed metal fuse with a central void gives a more reproducible, higher final resistance than blown fuses, which is the property that matters for sensing yield. Reference: C. Kothandaraman, S. K. Iyer and S. S. Iyer, ‘Electrically Programmable Fuse (eFUSE) Using Electromigration in Silicides’, IEEE Electron Device Letters 23(9), 523-525 (2002) (final resistance garbled in the text) (fetch-tool summary); IBM (now GlobalFoundries), ‘Electrically programmable metal fuse’, US Patent 8,421,186 B2 (2013) (fetch-tool summary of the first 100,000 characters).

Ques: What field and time are needed to program a gate-oxide antifuse?

Sidense quotes about 6 $V$ for a 2 $nm$ oxide, 30 $MV/cm$ (computed), with breakdown typically in 100 $μ s$ or less and about 100 $μ A$ per 100 $n m^{2}$ forming the conductive channel, a spot current density of about $1 \times 1 0^{8}$ $A/c m^{2}$ (computed). US 7,009,891 describes an antifuse capacitor of about 20 $\overset{˚}{A}$ breaking down at about 4.6 $V$ , 23 $MV/cm$ (computed), and programmed with 6 $V$ effective by splitting it between a 2.5 $V$ gate and a -3.5 $V$ well, so that neither rail exceeds what the transistors can withstand. Han et al. saw breakdown near 4.4 $V$ and programmed at 5.0 $V$ . In all cases the field is far above normal operation, so the breakdown is fast, but the programming power must be limited to avoid voiding the oxide. Reference: J. Lipman and T. Humes (Sidense), ‘A Flexible, Field-programmable ROM Replacement’, Design and Reuse (15 March 2010) (fetch-tool summary); ‘System and method for one-time programmed memory through direct-tunneling oxide breakdown’, US Patent 7,009,891 (assignee, inventors and year not in the text read) (fetch-tool summary); J.-W. Han, D.-I. Moon and M. Meyyappan (NASA Ames), ‘One Time Programmable Antifuse Memory Based on Bulk Junctionless Transistor’ (NASA NTRS 20190002597; year not printed, about 2018) (fetch-tool summary).

Ques: Is the resistance of a programmed antifuse hundreds of ohms, tens of kilohms or about a megohm?

All three are reported, for different cells and read paths. US 7,009,891 claims hundreds of ohms or less for its direct-tunnelling antifuse, against a few kilohms to 10 to 100 $kΩ$ for conventional antifuses. Jin et al. design their 0.18 $μ m$ BCD OTP for a post-program resistance of several tens of kilohms. Han et al. read about 1 $μ A$ at 1 $V$ through a junctionless cell, an effective 1 $MΩ$ including the junction (computed). The filament resistance depends on the power allowed during breakdown: a hard breakdown at high compliance gives low resistance, a soft breakdown a high one (background knowledge). For a sense-amplifier design, tens of kilohms is the more likely worst-case value to plan for, with hundreds of ohms an optimistic hard-breakdown value (my reading). Reference: ‘System and method for one-time programmed memory through direct-tunneling oxide breakdown’, US Patent 7,009,891 (assignee, inventors and year not in the text read) (fetch-tool summary); L. Jin, J.-H. Jang, D.-H. Kim, P.-B. Ha and Y.-H. Kim, ‘Design of 1 kbit antifuse one time programmable memory IP using dual program voltage’, Journal of Central South University 18(1), 125-132 (2011), doi:10.1007/s11771-011-0669-7 (abstract only); J.-W. Han, D.-I. Moon and M. Meyyappan (NASA Ames), ‘One Time Programmable Antifuse Memory Based on Bulk Junctionless Transistor’ (NASA NTRS 20190002597; year not printed, about 2018) (fetch-tool summary).

Ques: Why must unselected antifuse cells be protected during programming?

Because oxide breakdown is a cumulative wear-out process, and unselected cells on a shared line see part of the programming voltage. In Han et al.’s array, unselected cells see 2.5 $V$ in 60 $μ s$ pulses, which raised their unprogrammed current by up to ten times through stress-induced leakage, and the operating conditions are rated for 10^{3} write cycles. With a power-law voltage acceleration and an assumed exponent of 40, a cell that breaks down in 100 $μ s$ at 5 $V$ would last about $1.1 \times 1 0^{8}$ $s$ , 3.5 years, at 2.5 $V$ (computed), so halving the voltage gives a large but finite margin. Splitting the programming voltage between gate and well, as in US 7,009,891, keeps each node lower, and inhibit biasing of unselected lines does the same. Reference: J.-W. Han, D.-I. Moon and M. Meyyappan (NASA Ames), ‘One Time Programmable Antifuse Memory Based on Bulk Junctionless Transistor’ (NASA NTRS 20190002597; year not printed, about 2018) (fetch-tool summary); ‘System and method for one-time programmed memory through direct-tunneling oxide breakdown’, US Patent 7,009,891 (assignee, inventors and year not in the text read) (fetch-tool summary); Electrical Stress page of this vault (Reliability and Degradation, Stress Types).

Ques: When is an antifuse preferred over an eFuse?

When area, programming current or security matter most. Sidense notes that a 1T split-channel antifuse cell is not much larger than a mask-ROM cell, reads in as little as 10 $ns$ at advanced nodes, and cannot practically be read by de-processing or voltage scanning, because no charge or visible link change marks the bit. An eFuse needs a programming transistor that can carry about 10 $mA$ , as in Kothandaraman et al., which makes each bit large, and a programmed silicide gap can be seen in cross-section (background knowledge). On the other hand, eFuses are built from the standard gate stack or metal with no added steps and give megohm-level changes that simple circuits can sense, so they are widely used for redundancy repair, while antifuses need about 5 to 6 $V$ from a charge pump or external supply. Reference: J. Lipman and T. Humes (Sidense), ‘A Flexible, Field-programmable ROM Replacement’, Design and Reuse (15 March 2010) (fetch-tool summary); C. Kothandaraman, S. K. Iyer and S. S. Iyer, ‘Electrically Programmable Fuse (eFUSE) Using Electromigration in Silicides’, IEEE Electron Device Letters 23(9), 523-525 (2002) (final resistance garbled in the text) (fetch-tool summary).

Quartz 5

Explorer

Electrical Fuses and Antifuses