February_EDFA

edfas.org

ELECTRONIC DEVICE FAILURE ANALYSIS | VOLUME 18 NO. 1

well. By the end of the year, Dr. Boit expects to have more

experiments completed using a Hamamatsu detector for

modulation results. Therewas further audience discussion

centered around the power needed for the laser-assisted

device alteration (LADA) effect, the detected waveforms

that had spikelike artifacts on the leading edge, and finally

why Dr. Boit included e-beamtechnology in his 2019 devel-

opment roadmap.

Dr. Baohua Niu (Intel Corp.) gave the second presenta-

tion, “1154 nmLaser for Defect Localization, Design Debug

and FI Applications.” Current optical techniques in the

near infrared use heavily both the 1064 nm and 1300

to >1340 nm wavelengths for debug applications. The

1064 nm gives the highest resolution, while lasers in the

1300 nm wavelength range are losing steam in the indus-

try due to the aggressive process shrink and the required

resolution needed to effect productive node isolation, and

so on. Using the 1064 nmwavelength also has some draw-

backs. While it is the industry’s workhorse wavelength for

resolution and optical techniques, the SIL materials used

to support it have different properties. While the 1064 nm

wavelength, used inconjunctionwith the siliconSIL, has the

best possible imaging resolution, for higher-temperature

work, the SIL becomes highly absorptive and less stable

to work with over the range of the silicon area at differ-

ent operating temperatures. The 1064 nm wavelength,

when used with GaAs material, does not have the same

temperature instability as the above but has slightly less

resolution versus the aforementioned silicon SIL material,

has the propensity to corrode over time, and is expensive.

Due to the above issues for both the wavelength and SIL

materials, Dr. Niu explored to see if there was an alterna-

tive to the aforementioned issues by using a different

wavelength. The 1064 nm laser is below the 1107 nmsilicon

indirect bandgap, and the 1319/1340 nm lasers are above.

Becauseminimumabsorption occurs at 1154 nm in doped

siliconand is ~20

lower than that of 1064nm, thequestion

becomes: Could it also be a good light source for optical

techniques that we use today as an alternative, and, if so,

what are the tradeoffs? Dr. Niu showed the experimental

results using a system with an 1154 nm light source and

a silicon SIL tip with a 3 mm radius with <30% signal loss.

Due to advantages in the bandgap of silicon, the 1154 nm

laser hasmuch less effect on silicon versus 1064 nm(charge

injection). A number of real data examples were discussed,

includingwaveforms and continuous-wave signal imaging

and probing (frequency maps), showing how well the

1154 nm laser was used. The LADA effect for 1154 nm

is similar to that of 1064 nm (typical shift of 5 to 10 ps).

The modulation signal strength is ~2

stronger with

1154 nm versus the current 1064 nm optical techniques.

For waveform probing, the transistor signal strengths

are comparable; PMOS has the same strength as NMOS

devices. Although with the higher 1154 nm wavelength

(versus 1065 nm) there is a small 5% loss in overall

resolution, the strongmodulationof light andother consid-

erations that make the 1154 nm an acceptable alternative

to the 1064 and 1300 nm technologies can further simplify

the overall optical probe tool configuration by use of only

one wavelength (1154 nm) versus the dual-wavelength

system, comprised of both the 1064 and 1300 nm lasers.

The cost of SIL tip material and overall tool ownership can

also be reduced.

Dr. Joy Liao (NVidia Corp.) gave the third presenta-

tion, “Volume Electrical FA for Product-Specific Yield

Enhancement.” Dr. Liao started her presentation by

stating that foundry-based companies cannot ramp yield

alone. She also made strong observations that yields are

increasingly product-specific and that DFTmethodologies

are key to ramping yield, including shift, chain, automatic

test pattern generation (ATPG) at speed, memory built-in

self-test, and so on. Physical failure analysis, based on

diagnosis alone, has poor success rates. Shift failures are

a huge portion of early failures. Reliability failures often

appear as shift failures aswell. Dr. Liaoalso talkedabout the

application of electrical failure analysis (EFA) techniques to

observe internal circuit behavior. She showed a number of

test cases, including:

• Good/bad die scan chain image/comparison, using

emission toolset/techniques

• Modulation mapping on a broken scan chain

• Modulationmappingandcontinuous-wave laser voltage

probing (LVP) on a broken scan chain

• Combination of both LVP and emission on a broken

scan chain

• Soft defect localization (SDL) and ATPG at-speed

diagnosis

• SDL with voltage dependence

Dr. Liao further discussed the capability to automate

chain navigation with softwaremeans tomakemore intel-

ligent localization decisions. On 20 nm, more aggressive

DFT and test compression is challenging. Moving forward,

more aggressive DFT architecture continues to be needed.

Less data are available from production wafer-level

testing, further limiting volume software diagnosis. More

complicated test loops and requirements for LVP work

are becoming more challenging as well. Up to five broken

scan chains per day were solved. The combination of LVP,

modulation mapping, SDL, and emission techniques is

S - 3