S HORT AND L ONG -TERM S AFE O PERATING A REA C ONSIDERATIONS IN LDMOS T RANSISTORS
英语a级试题
Philip L Hower and Sameer Pendharkar* Mixed-Signal Technology Development
Texas Instruments
Manchester, NH 03101, and *Dallas, TX 75321, USA
[Keywords: LDMOS, SOA, hot carrier lifetime, thermal instability, WLR, ESD, CHC lifetime, thermal SOA, electrical SOA]
I NTRODUCTION
Lateral DMOS transistors are widely ud in mixed-signal IC circuit designs, particularly where power handling is important. Both high-voltage (20 to 100 V) and high current (2 to 3A) output drivers are required for applications within the automotive, paper media, digital audio, and other markets. This paper views the LDMOS fromresist
a power-handling perspective, considering both design and characterization aspects.
Safe operating area regimes
In designing and using the LDMOS, particular attention must be paid to the electrical stress placed on this component. If the stress exceeds certain limits, the device is likely to fail. The phra “safe operating area” or SOA is ud to define the boundaries that limit the device operating point to conditions that prevent device failure. Becau devices can be stresd in a variety of ways, SOA definition is not straightforward. To help with SOA determination, it makes n to define not one, but veral safe operating areas, distinguishing them mainly by the duration time of the electrical stress. Three different SOAs are indicated in the plot of Figure 1, which shows stress time as the horizontal axis. Electrical and thermal SOAs typically deal with stress times less than a few ms, that is, their boundaries are determined by “short-term” testing. Alternatively, hot carrier stress tests typic
ally require times from a few hundred to veral thousand conds and are labeled “long-term.” An important point is that the maximum allowed power density, shown as the vertical axis in Figure 1, depends on stress time and has the general shape indicated in Figure 1.
ruhestress time (s)
caroline什么意思P /A (W /c m 2)
F IGURE 1. M AXIMUM POWER DENSITY VS . STRESS TIME FOR A TYPICAL LDMOS, SHOWIN
G VARIOUS SOA REGIMES .
Short-term vs. long-term SOA
Short-term SOA regimes can be linked to two different carrier generation process that trigger catastrophic device failure [1]. Both impact ionization and thermal generation can create “excess” carriers within the device. The carriers are labeled excess becau they provide currents that are not under gate control. If the excess currents reach levels comparable to the current being controlled by the gate, instability and device failure can occur.
Rather than detecting catastrophic device failure, hot carrier stress
tests monitor degradation of some fundamental device parameter, for example, on-resistance, Rdson, or threshold voltage. The degradation is associated with trapping of impact generated carriers at the oxide/Si interface under the gate or within the drift region [2], [3],
[4]. Impact ionization dominates electrical and HC stress safe operating areas and can play a condary role in thermal SOA. Thermal generation is mainly associated with thermal SOA. H IGH -VOLTAGE MOS TRANSISTORS
scum
Conventional MOS transistors are generally not uful for processing power. They are designed inst
ead to process information (bits) using minimum input energy while occupying the least area. Gate lengths are minimized as is gate oxide thickness. Figure 2 shows a simplified cross-ction of a conventional NMOS transistor.
F IGURE 2. NMOS CROSS -SECTION WITH APPLIED V DS AND WITH V GS =0. T HE DRAIN DEPLETION LAYER IS INDICATED BY THE DASHED LINES .
Maximum drain voltage is limited by gate oxide breakdown as well as by the combined effects of punch-through and avalanche breakdown from drain-to-source. CMOS technologies currently have maximum drain voltages moving into the 1 to 2 V range. This means that changes need to be made to conventional NMOS and PMOS geometries if they are to handle power.
Extending the drain
By interposing a lightly-doped n-type gap between drain and gate, the drain voltage can be incread. The idea is to avoid exceeding the silicon critical electric field, Ecrit, while increasing drain-gate distance just enough to reach the required voltage. Ecrit ranges from 10 to 30 V/µm and depends on how well parasitic npn action is suppresd [1]. Figure 3 illustrates a “drain extended” NMOS or DENMOS. In many cas, the positive charge in the depleted n-region can be balanced by
the negative charge in the p- layer underneath. With this approach, “RESURF” (reduced surface field) [5], [6] action becomes possible. The electric field becomes nearly horizontal and fairly uniform along the low-doped region. By maintaining charge balance and using the appropriate vertical profiles, the maximum Vds can be incread simply by increasing the length of the low-doped region. One negative conquence of this approach is incread Rdson. The drain Nwell can usually be ud to provide the drain extension, but in some cas an extra mask may be needed.
F IGURE 3.D RAIN-EXTENDED NMOS(DENMOS) CROSS-
SECTION.T HE ENTIRE LOW-DOPED REGION IS DEPLETED.
The Lateral DMOS
One difficulty with the DENMOS is that the gate length needs to be fairly long to avoid punchthrough.
This is due to the relatively light doping of the p-region under the gate. By adding a mask and implant step that increas this doping to form a laterally diffud “body” region beneath the gate, one can create a lateral DMOS or LDMOS. The LDMOS can have channel lengths an order of magnitude less than a DENMOS. This means that it is possible to significantly reduce device area, while achieving low channel resistance and a corresponding favorable increa in trans-conductance. The penalty is a moderate increa in threshold voltage, which is usually acceptable in power applications.
Figure 4 shows the cross-ction of one style of lateral DMOS. In this ca the gate and drift region are both confined to the thin oxide or active region, forming an “active-gap” LDMOS.
A cond style of LDMOS, which has a “field-gap,” is shown in Figure 5. In this ca, the gate is extended over the thick field-oxide, thereby providing a field-plate that reduces junction curvature effects. The field-gap style is preferred for voltage ratings above 20 to 30V.
Figure 6 summarizes the main features of the DEMOS and LDMOS. In general, for high-voltage, low-current applications, the DEMOS, is preferred. Where both high-voltage and high-current are required, the LDMOS is the preferred device. For this reason, it is usually the LDMOS that leads to questions about safe operating area, which is discusd in the next ction.
F IGURE 4.C ROSS-SECTION OF AN A CTIVE-GAP LATERAL
DMOS.T HE P-BODY IS SELF-
ALIGNED TO THE SOURCE OR GATE EDGE.
F IGURE 5.C ROSS-SECTION OF A FIELD-GAP LDMOS.T HE GATE OVERLAPS THICK FIELD OXIDE.
disgracefulF IGURE 6.C OMPARISON OF DRAIN EXTENDED MOS
TRANSISTORS AND LATERAL DIFFUSED MOS TRANSISTORS.
T HE OBSERVATIONS APPLY TO BOTH P AND N-CHANNEL.
S HORT-T ERM LDMOS S AFE O PERATING A REA
Undesired turn-on of the parasitic bipolar that exists in parallel with the LDMOS can lead to the device suddenly switching to a low-impedance state. Although the triggering of device failure is an electro-thermal event, it is usually the ca that one mechanism dominates. Therefore it is convenient to think of the overall SOA as being the combination of a thermal SOA and an electrical SOA [1]. SOA boundary conditions
The electrical SOA boundary is where the device begins to exhibit negative resistance. The resulting “snapback” to a low-voltage state is rapid and occurs within tens of pico-conds. In some
cas, the LDMOS is able to survive the transition to the low-impedance state [7], however, the snapback event is disruptive to circuit operation. Therefore the safe operating boundary is defined to be at the ont of negative resistance. Within the device, electrical SOA can be characterized by the
field at the drain contact reaching a critical value, Ecrit . From the standpoint of terminal conditions, electrical SOA is best reprented by a curve in the Id, Vds plane. Electrical SOA is discusd further in a later sub-ction.
The thermal SOA boundary is defined by the device being on the verge of thermal instability or thermal runaway, where the thermal instability is brought about by LDMOS lf-heating. In some cas, LDMOS lf-heating can be assisted by impact generated drain current [8], however, the triggering mechanism is still dominated by thermal instability. The thermal SOA boundary can be characterized by the junction reaching a critical temperature, Tcrit . As described in the next sub-ct
ion, this temperature can be extracted from measured power-pul failure points using a model of device thermal resistance that corresponds to the measurement pul time. With Tcrit and thermal resistance Rth(tp) known, any power pul can be checked to e whether it exceeds the thermal SOA. Determining thermal SOA
Figure 7 shows electro-thermal simulation results for temperature and drain current per unit gate width, for a 60V LDMOS. A power pul of Vds=50V and Vgs=4V is applied. Initially, the junction temperature shows the usual parabolic respon associated with thermal diffusion. After about 80µs, the device exhibits thermal runaway of drain current at a junction temperature near 800K. Plea note that at the beginning of the pul, the drain current temperature coefficient is negative, corresponding to a thermally stable situation. Nevertheless, as time (and temperature) increas, a point is reached where the temperature coefficient changes sign and becomes positive, leading eventually to thermal instability.
T J (K )
300400
500600700
demonstrate
800900I d /W (A /µm )
F IGURE 7. S IMULATED JUNCTION TEMPERATURE AND
NORMALIZED DRAIN CURRENT VS . TIME IN RESPONSE TO AN
APPLIED POWER PULSE .
Figure 8 shows a plot of “measured” Tcrit vs. drain-source
voltage for LDMOS from two different power IC technologies. For
the measurements, low duty cycle Vds puls (1ms) are applied, and Vgs is gradually incread until thermal runaway is detected at the end of the pul. For example, e [9] for a test circuit and waveform examples. For a given pul, Tcrit is calculated using a thermal model of the device, [8],[10]. As shown in the plots of Figure 8, there is some variation of Tcrit with Vds and also between the two types. For large Vds , impact ionization provides carriers (holes) that act as ba drive for the parasitic npn, enhancing npn turn-on, and thereby reducing Tcrit . This effect is less pronounced in the Type B LDMOS due to improved shorting of the ba-emitter junction of the parasitic npn. At lower Vds, there is a notable increa in Tcrit for both LDMOS. Bad on preliminary simulation studi
es, this increa appears to be due to the ballasting effect of npn emitter/source resistance. Further work is needed to fully understand the shape of the Tcrit curve at low Vds .
Independent of the interesting physical explanations, the Tcrit vs. Vds curve and the LDMOS thermal model provide uful tools for predicting thermal SOA without the need of building prototype devices. The advantage of using Tcrit for device design is that it is esntially independent of device size and pul time.
F IGURE 8. CRITICAL TEMPERATURE VS . DRAIN SOURCE VOLTAGE FOR 50V L DMOS FOR TWO DIFFERENT DESIGNS . T CRIT IS EXTRACTED FROM 1 ms PULSE MEASUREMENTS .
Electrical SOA
Figure 9 depicts electron flow within the drift region. The situation can be approximated by assuming that there is a pudo one-dimensional “flux tube” within the drift region. Electrical instability can ari when both Vds and current density within the drift region are sufficiently high [11].
When the current density exceeds a value given by the average
doping in the drift region, negative charge from the electrons shifts
the slope of the field profile within the flux tube. This action is illustrated at the top of Figure 9, where the dashed line shows the field profile at low currents. The solid line shows the profile for the
ca where the current is large enough to tilt the profile and where
Ecrit is reached at the drain contact. In the latter state, increasing
current density leads to decreasing voltage and the device is in a
negative resistance or snapback mode. Electrical SOA limits apply,
provided temperatures do not ri to the point of initiating thermal
instability. Electrical SOA is appropriate for an ESD HBM event
(~100ns puls) or during rever recovery of a body diode in an H-bridge, where recovery times are typically in tens of nano-conds.
Examples of measured electrical SOA are shown in Figure 10. For the measurements a transmission line pul technique (TLP) is ud, where puls are applied drain-to-source and Vgs i
s held constant [1]. The pul time is 100ns. Each point on the curve corresponds to the ont of snapback for a particular value of Vgs. It can be en that there is a major improvement in e-SOA for the Type B LDMOS. This increa is directly related to the lower ba-emitter resistance shunting the npn, which permits an effective increa in Ecrit [12]. The on-resistance, area product (Rsp ) of the two LDMOS is about the same.
E-field
Ecrit
F IGURE 9. LDMOS CROSS -SECTION SHOWIN
G THE FLOW
PATH FOR CONVENTIONAL CURRENT AND FOR ELECTRONS AND HOLES WITHIN THE DRIFT REGION . T HE DASHED FIELD PROFILE IS AT LOW CURRENT DENSITY AND THE SOLID IS AT HIGH CURRENT DENSITY .
F IGURE 10. M EASURED SNAPBACK DRAIN CURRENT VS . DRAIN VOLTAGE AS OBTAINED FOR DIFFERENT V GS . T HE CURVES DEFINE THE ELECTRICAL SOA FOR TWO DIFFERENT 50V LDMOS. D RAIN CURRENT IS NORMALIZED TO GATE WIDTH .
In the plots of Figure 10, drain current is normalized to gate width, W . From experimental measurements on devices having a wide range of gate widths, we know that the curves are esntially independent of W. Thus data similar to that of Figure 10 can be ud to scale devices to meet a particular electrical SOA requirement. Again, the need for building prototype devices is substantially reduced by using this type of curve.
L ONG -T ERM LDMOS S AFE O PERATING A REA
Since current flow and impact generation in LDMOS devices can be clo to the miconductor-dielectric interface, high electric fields can impart sufficient energy to the carriers so that they overcome the energy barrier at the interface and get injected into the dielectric. This hot carrier injection can significantly affect the device performance. Although it is not a catastrophic failure, device operation is limited to prevent any significant degradation due to hot carrier injection (HCI). The operating regime so defined is called hot carrier stress SOA (HCSOA). Hot carrier stress SOA
HCSOA for an LDMOS device is generally limited by degradation in on-state resistance for high drain and low to medium gate voltage conditions and by degradation in threshold voltage for high gate voltage bias conditions.
Degradation in the threshold voltage (Vt ) depends upon the resistance of the source link, which is the lightly doped region connecting the channel to the heavier doped source region. As shown in [4], Vt degradation can be substantially improved by using a lower resistance source link. Such a low resistance source link may not be always compatible with the lightly doped regions ud in NMOS devices and may need additional processing. Rdson degradation
Degradation mechanisms affecting the on-state resistance of the LDMOS are addresd in greater detail in this sub-ction. Figure 11 plots the impact ionization contours and potential distribution for an active-gap LDMOS device for “worst” ca stress condition. The worst ca stress condition is defined as the bias condition causing the most degradation in Rdson . As en from Figure 11, the high electric field near the surface caus significant impact generation clo to the interface.
F IGURE 11. I MPACT IONIZATION CONTOURS AND
POTENTIAL DISTRIBUTION FOR WORST CASE STRESS CONDITION .
Two dimensional simulations of the device structure, including traps in the dielectric, show that the obrved degradation in Rdson is
due to the electron capture by traps in the dielectric area over the drift region but clo to the gate oxide area [2]. This is shown in Figure 12 below. It is propod in [3] that the traps are generated due to hot hole injection when the device is operating at high voltage and high current conditions.
Since the Rdson degradation primarily occurs due to hot carrier injection near the gate-edge – drift-region boundary, the HCSOA of the LDMOS is critically dependent on the drain engineering of the device. The doping profile in the drift region and particularly in the region under the thin gate oxide has a significant influence on the HCSOA. Optimizing this profile can substantially reduce the hot carrier injection and associated degradation and improve the overall HCSOA as shown in Figure 13 below.
F IGURE
12. E LECTRON DISTRIBUTION , CURRENT
FLOWLINES AND TRAPPED CHARGE DISTRIBUTION SHOWING ~5% DEGRADATION IN RDS ON AFTER 3000 SECOND STRESS TIME .
exoticMeasured and simulated Idlin degradation for different ldmos devices
Stress time (conds)
% d e c r e a s e i n I d l i n
F IGURE 13. M EASURED AND SIMULATED ID LIN DEGRADATION FOR
LDMOS DEVICES WITH DIFFERENT DRAIN ENGINEERING . T HE T YPE I DEVICE SHOWS S
68届金球奖ATURATION IN DEGRADATION (<10% DEGRADATION FOR 10 YEAR OPERATION . (ID LIN IS MEASURED AT V DS =0.1V AND IS EQUIVALENT TO MEASURING R
DSON ). The plots of Figure 13 show that the Type I device gives much less degradation than the Type II device and therefore has a superior HCSOA. The main difference between the two types is that for Type II, the drain extension is a shallow implant inside Nwell, while Type I us a deeper drain extension [2]. The deeper drain extension in the Type I device diffus significantly more under the thin gate oxide region. This not only reduces the number of injected carriers in the drift region but also allows the channel electrons in Type I device to spread out deeper as they enter the drift region. The deeper current flow results in less influence of the injected carriers causing a much lower degradation in Type I on-state resistance.
It should be noted that although the data for an active-gap 16V rated LDMOS is shown here, the device behavior and degradation mechanisms are generally similar for field-gap LDMOS devices.
O VERALL LDMOS S AFE O PERATING A REA
CHARACTERIZATION
gripe
In order to minimize the intrinsic ba resistance and improve the device robustness, LDMOS devices normally u “integrated” source / p-body contact strategy. For this reason, the body current cannot be directly measured in the devices. This, coupled with the fact that the hot carrier degradation in the devices does not follow the same “degradation vs. Isub” power law relationship as in a standard NMOS, means that HCSOA characterization has to be done differently for LDMOS devices. The HCSOA characterization typically includes stressing the device at multiple drain and gate voltages and then extrapolating to find the time for degradation by a predetermined amount (usually 10% of the unstresd value). This data can then be combined along with electrical SOA characterization data to create a SOA plot in the Vds – Vgs plane.
Figure 14 and Figure 15 show combined SOA plots for the two types of LDMOS shown in the previous ction. Different color coding or shading is ud to identify the regions suitable for analog and digital operation. Digital operating region is then further sub divided depending upon allowed duty cycle. The overall improved SOA of the Type B LDMOS device can be readily en from the plots.
F IGURE 14. T YPICAL SOA PLOT FOR A T YPE A LDMOS DEVICE COMBININ
G HCSOA AND E-SOA . THE ALLOWED DUTY CYCLE PROGRESSIVELY DECREASES GOING FROM LEFT TO RIGHT . T HE DEVICE FAILS CATASTROPHICALLY IF OPERATED IN THE REGION DENOTED AS “ELECTRICALLY UNSAFE ”
